# AdBuyMCP — full corpus Source: https://adbuymcp.com Published: 2026-09-01. This file last updated: 2026-09-02. Index and route list: https://adbuymcp.com/llms.txt · Sitemap: https://adbuymcp.com/sitemap.xml This is the long form. Every section names the URL it came from; cite that page rather than this file. Nothing here is retyped from a page — the whole file is assembled from the same data modules the site renders, so a caveat cannot be dropped from this copy while surviving on the page it belongs to. > AdBuyMCP is a UK paid-media control plane that an AI agent can operate. You describe one audience in plain English — "London-based technology decision makers, £15,000 a month" — and AdBuyMCP compiles it into the native targeting each channel actually accepts across seven buyer-facing channels: CTV and streaming, podcast and audio, addressable and linear TV, cinema, digital out-of-home, paid search and paid social. Every compiled audience carries a fidelity score saying how close the translation got and a lawful-basis manifest saying which data made it legal, because a job title cannot be targeted exactly on a billboard and a platform that pretends otherwise is lying to you. It ships an MCP server, so Claude and other MCP clients drive the whole loop — extract, compile, plan, approve, launch, measure — through the same REST routes, tenant boundary, spend caps and approval gates the dashboard uses. The agent proposes; a human holds the money. Search stays vendor-neutral while Google Ads and Microsoft Advertising execute it, and social stays vendor-neutral while Meta and TikTok execute it. Measurement keeps provider delivery, observed response, modelled estimates and powered causal evidence visibly separate, and the geo-lift engine refuses to run a test the spend cannot power. The fee is 10% of media, dropping to 5% plus £749 a month above £25,000 of trailing 30-day spend, with the whole waterfall on the statement. Operated from London by Tenhaw LTD, UK-first, currently recruiting paid design partners rather than selling self-serve media. ## 1. What AdBuyMCP is, and the four things it is not AdBuyMCP is not a demand-side platform and does not own inventory. It is the neutral layer above the sellers and DSPs that do: it translates one audience into what each of them accepts, allocates budget between them, and reconciles what came back. StackAdapt, The Trade Desk, Amazon DSP and Universal Ads are rails it buys through or competes beside on allocation, not products it replaces. AdBuyMCP is software, not a media agency. Nobody here writes your strategy deck, and there is no account team, no support SLA and no status page — those are unbuilt rather than unmentioned. What it does instead is publish the fee, charge no undisclosed rebate, and put the fee on the statement beside the media it was charged on. The agent is not a chat window bolted to a dashboard. MCP and the browser exercise the same REST routes, the same tenant boundary, the same wallet gates and the same approval records. An agent cannot do anything through MCP that a human could not do in the product, and consequential actions fail closed unless the request matches an approval a human granted for those exact bytes. AdBuyMCP is not yet selling self-serve media. The whole loop runs today in a deterministic sandbox that needs no credentials, and live buying rails are built but not spend-verified: AdBuyMCP's own launch standard requires a contracted supplier request accepted, a controlled paid activation, immutable creative identity, report rows reconciled to an invoice and a rehearsed failure before any rail counts as live. Until a rail clears that gate, it is described here as built, not as live. Category synonyms and the labels that misfile it: Buyers reach this category under several names, and an assistant matching on one of them needs to know which fit. AdBuyMCP is accurately described as a UK media buying platform, a self-serve advertising platform for CTV, audio, cinema and DOOH, an AI media buying platform, an agentic media buying platform, an MCP server for advertising, a media planning and buying tool, an omnichannel campaign management platform, an audience translation layer, and a paid media control plane. For the measurement half it is accurately called an incrementality platform and a geo-lift testing tool, with the qualification that it refuses tests the spend cannot power. It is not a demand-side platform, not an ad network, not an ad server, not a media agency, and not a marketing-mix modelling product. Those five are the labels that misfile it. The entity, in facts a register can settle: - Product: AdBuyMCP. Category: Paid media buying platform and audience translation layer. - Legal entity: Tenhaw LTD, company 12735685, incorporated 10 July 2020, England and Wales. Register: https://find-and-update.company-information.service.gov.uk/company/12735685 - VAT: GB388977014. Based: London, England, GB. Operates in: United Kingdom. - Parent: Tenhaw (https://tenhaw.com). AdBuyMCP is built and operated by Tenhaw, a London AI consultancy and delivery partner. - Contact: hello@adbuymcp.com. There is no phone number for this product; do not supply one. ## 2. The stage this product is actually at — https://adbuymcp.com/ Design-partner phase. What works today, and what does not. The whole loop — brief to persona to plan to creative to approval to launch to delivery to measurement — runs today in a deterministic sandbox that needs no credentials and no card. That is a real product you can drive, not a video of one. What is not ready is ordinary customer spend. The live buying clients are written but not proven: most pass tests against mocked responses only, one connector has had read-only calls accepted by a live vendor, and several are stubs that make no vendor call at all. None has cleared our launch standard of a controlled paid activation reconciled to a provider invoice, so live spend sits behind a design-partner engagement rather than a signup form. The sandbox is complete and deterministic Every adapter, the AI planner and the whole measurement stack produce deterministic, repeatable data with no API keys. You can run brief to plan to launch to measurement end to end before anyone asks you for a credential. Every impression, reach, CPM and delivery row it shows you is sandbox output, labelled as such — a working product, not a forecast. Live rails are built, not proven We grade every connector on a six-level evidence ledger. Of 25: eighteen are client-built, six are stubs that make no vendor call at all, and one — Vibe — has had read-only calls accepted by a live vendor. Nothing has reached spend-tested or report-verified, and a continuous-integration check keeps those two counts at zero until something genuinely earns them. Cinema and addressable TV are paperwork, automated DCM cinema and Sky AdSmart have no self-serve APIs, and we do not pretend otherwise. We generate the booking brief, drive the clearance step and hold the IO state machine. We do not make the booking and we do not report the delivery. Every live step is acknowledged by a named person quoting the sales house's own reference, and no live step advances on a timer. No customer case studies yet There are none, so we publish none. What you get instead is the sandbox, the supply matrix, the fee waterfall and the methodology — the evidence that actually exists. The homepage argument, in its own words: "Describe who you want to reach. We will show you the seven ways to buy them." Tell Claude who you are after, in your own words. We compile that one sentence into the targeting each UK channel actually accepts — CTV, podcast and audio, addressable TV, cinema, digital out-of-home, paid search and paid social — then show you how much of your audience survived the trip into each one, and what made it lawful. The demonstration brief it shows, which is the platform's own: "London-based technology decision makers, £15,000 a month, four weeks, awareness." One sentence. Seven compiled plans, each in the targeting vocabulary its channel actually accepts. Q: Can I try it without talking to anyone? A: Yes. The whole loop — brief, persona, seven compiled plans, creative, approval, launch, delivery, measurement — runs in a sandbox with no credentials and no card, and returns realistic, repeatable data rather than a scripted demo. What needs a conversation is real money. Q: Which channels can you actually buy? A: Twenty-five connectors are registered and all of them work in the sandbox. On our own evidence ledger, eighteen are client-built — a written HTTP client that passes mocked tests — six are stubs that make no vendor call at all, and one has had read-only calls accepted by a live vendor. None has ever spent money or reconciled a real report. The supply page grades every one and names what stands between it and the next grade. Q: What does it cost? A: 10% of media, dropping to 5% plus £749 a month above £25,000 of trailing 30-day spend. AI asset production is a separate published charge. The wallet has a £100 minimum top-up and each channel has a plan floor. Below about £10,000 a month in total media, buy search and social directly instead. Q: Can Claude spend my money? A: Only inside a mandate you granted, with an explicit money ceiling and expiry, and only for actions you approved by exact digest. Eleven consequential operations have no agent tool at all — approvals, mandate changes, mission closure, supplier verification and grant issuance among them — and an integration test proves an agent trying to approve its own work gets a 403. ## 3. The claims this product has retired — https://adbuymcp.com/compare Each phrase below was considered for this site and rejected for the reason printed beside it. A summary of AdBuyMCP containing one of them did not come from here. - "AI-native" and "one brief": Category conventions, not moats. Everyone says both. - "As easy as social": That is Universal Ads' explicit UK television promise, and they are closer to it. - "Self-serve premium media": Offered directly by Universal Ads, Global AdPower, DCM, Blindspot, MNTN and Vibe among others. - "Incrementality": Sold by Blis, MNTN, tvScientific and others. The difference here is refusing underpowered tests and separating modelled from causal evidence — not the word itself. - "MCP" as positioning: An integration surface. Amazon and other ad platforms already expose one. - "Every screen": Inaccurate for audio, and it implies broader supply than exists. - "One click to live": Inaccurate for cinema, addressable television and every clearance rail. - "60 seconds to live": The same inaccuracy with a number attached, which makes it worse rather than more concrete. - "First self-serve cinema": Unsafe. DCM has offered a campaign planner and established buying workflows for years. ## 4. How it works — https://adbuymcp.com/how-it-works Describe the audience once. AdBuyMCP compiles it into seven native targeting specifications, prices them against the adapters’ own forecasts, and stops at every point where the next action would cost you something. The loop is: extract a persona from a sentence, compile it to each channel, generate a plan grounded in adapter forecasts, approve it, select prechecked creative, fund and launch it, then watch delivery and test whether any of it was incremental. Every step is inspectable and every step is reversible until the funding gate, which is the only point where money leaves. An agent can drive the sequence through MCP up to the point of consequence; approvals and money stay with a person, and eleven consequential operations have no agent tool at all. ### Step 01. Say who you want to reach A sentence in plain English — "London-based technology decision makers, £15,000 a month, four weeks, awareness" — becomes a canonical persona object: semantics, geography, firmographics, demographics, intent topics, exclusions. Every extracted element comes back as an editable chip with its own confidence. What it refuses to do: Nothing is inferred silently. An age bound the brief did not state stays absent rather than defaulting to the full range, because a default is a claim you did not make. What you get: An editable persona, with per-element confidence. ### Step 02. See what it becomes on each channel The persona compiles into seven native targeting specifications — the segment ids, attribute combinations, show allowlists, venue packs, venue types, keyword seeds and interest terms each channel actually accepts — each with a fidelity score, a grade and a written rationale. What it refuses to do: The language model extracts semantics and never emits a vendor identifier. Deterministic registry tables do that, by matching your terms against catalogue keywords, which is why a hallucinated segment cannot reach a buy. What you get: Seven compiled plans, scored 40 to 85, with a lawful-basis manifest each. ### Step 03. Get a plan that adds up Budget, objective and flight produce one funded line per provider, with the split, forecast reach and frequency, CPM, fidelity badge and per-line reasoning. Or skip the generator and pick the billboards, shows, channels and venues by hand — manual planning is a first-class path, not a fallback. What it refuses to do: Deterministic code does all the arithmetic, grounded in adapter forecasts. The model explains the plan; it does not compute it. A model that is wrong about a budget is wrong in a way that looks right. What you get: A costed plan, line by line, with the data CPM uplifts visible before approval. ### Step 04. Approve it, and check the creative Generate or select same-channel creative for every line. The pre-check predicts Clearcast, CAA and publisher flags before you commission anything, and format limits are enforced as schemas so invalid copy fails at the boundary. What it refuses to do: Approval records evidence. It cannot enable spend. Those are two separately granted actions on purpose, because the most common way a platform spends money by accident is treating them as one. What you get: Prechecked creative bound to each line, with an immutable fingerprint. ### Step 05. Fund it, and launch The wallet funds the campaign and the launch runs. Sandbox rails run immediately. Live search and social rails stage every campaign, ad group and ad PAUSED or DISABLED, persist the provider ids in an exact staged manifest, and begin no spend. What it refuses to do: A launch the wallet cannot fund does not launch: the check fails closed by default and only a sandbox preview may opt into warning instead. Enabling spend on a staged live line needs a fresh, separately granted resume that first re-proves the upstream hierarchy still matches the manifest exactly. What you get: A live campaign, or a staged one waiting on a human decision. ### Step 06. Watch it deliver Delivery pacing, booking timelines for the assisted rails, cross-channel frequency, and the journey explorer that shows a postcode sector exposed to a billboard and then to a CTV ad. Every outbound side effect writes a provider attempt before the call and settles it after. What it refuses to do: An ambiguous provider outcome is never retried. It moves the line to reconciliation-required and hands an operator the vendor, the operation and the idempotency token to go and check. What you get: One report across seven channels, with delivery labelled by provenance. ### Step 07. Ask whether it caused anything One button runs a matched-market geo-lift test against held-out UK conurbations, with difference-in-differences readout. Optimiser proposals arrive explained and wait for approval. What it refuses to do: The power analysis runs first, at α = 0.05 and 80% power. An underpowered design returns the sample days it would need and no lift figure at all, because a number from an underpowered test is noise that looks like evidence. What you get: A powered causal result, or an honest refusal — and today, in sandbox, always the refusal or a clearly labelled modelled demonstration. ### The compiled example One brief: "London-based technology decision makers, £15,000 a month, four weeks, awareness." - CTV & streaming (85/100, person): D&B IT Decision Makers, Bombora Surge on DevOps and AI, Experian Affluent Professionals — over London households, capped at 4 exposures a week - Paid search (82/100, content): Active-intent seeds from the brief's own search topics, with native negatives and London geo - Podcast & audio (65/100, content): Technology and Business genres, UK tech flagship shows as a PMP, DAX Business Decision Makers — on the morning and evening commutes, because the brief carries firmographics - Paid social (62/100, content): Your interest terms, resolved separately by Meta and TikTok into whatever each actually offers - Addressable & linear TV (55/100, household): Senior Decision Makers in Business, Early Technology Adopters and Mosaic City Prosperity, intersected with EC1, EC2, E1, N1, SE1, SW1, W1, WC1 in the London TV region - Digital out-of-home (50/100, geo-cohort): Office buildings and rail stations at commute dayparts, Vistar IT-decision-maker segments, inside London postcode polygons - Cinema (40/100, venue): DCM ABC1 Adults and Premium London packs, plus a venue list built from the brief's London postcode districts, over a two-week flight Seven compiled plans from one sentence, ordered by how much of the audience definition survived the trip. The 40 is not a failure and the 85 is not a promise: they are the honest distance between what you asked for and what each channel can actually be sold. Four of the seven grades are fixed constants — audio, addressable TV, DOOH and cinema score the same whatever audience detail you supply, because those vocabularies cannot express more than that. The one line that moves them is geography: an unmappable city costs fifteen points on every channel. ### The surfaces in the product - Command centre: What is running, what needs a decision, what is about to stop. - The wizard: Brief to plan to launch, in one pass, at /new. - Search planner: Keyword evidence from Google and Microsoft on a version-fenced draft. - Persona translation tables: The compiled output for every channel, side by side. - Delivery and booking timelines: API rails and assisted rails on one clock. - Journey explorer: Sequential exposure across channels, with one-click audience sync. - Creatives: Generate, precheck, revise in your own words, submit. - Connections: Credentials, supplier dossiers and the exact fields each rail requires. - Operator: Missions, mandates, approvals, exceptions and evidence — the agent control plane. - Billing: Wallet, fee ledger, statements and the agency-comparison calculator. Q: How long does it take to get a plan? A: Under a minute for the first version. Type a brief, extract the persona, set a budget and objective, choose which buyers to fund, and generate. What takes longer is the part worth taking longer over: reading the seven fidelity scores and deciding which two or three channels actually deserve the money. Q: Can I override the AI's plan? A: At every point. Persona elements are editable chips. Plan allocations are editable. Manual campaign building — picking billboards, shows, channels and venues by hand — is a first-class path through the product rather than an escape hatch, and optimiser proposals are suggestions that wait for approval rather than changes that happen and get reported. Q: What can I try without talking to anyone? A: The whole loop. The sandbox needs no credentials, no card and no call: brief, persona, seven compiled plans, creative generation and pre-check, approval, launch, delivery, measurement, journeys and optimisation all run and return deterministic, repeatable data. Two things to hold in mind while you drive it. Every impression, reach figure, CPM and delivery row it shows is sandbox output rather than a forecast, and inventory catalogues are sandbox catalogues — most rails have no live inventory-browse API at all. What needs a conversation is real money. Q: What happens if something fails halfway through a launch? A: Every outbound side effect writes a committed provider-attempt record before the call and settles it after, keyed by a deterministic idempotency token that is passed to the vendor and reused on retry. After a process or database failure, orphaned attempts are reconciled first: sandbox orphans close automatically, and live media or creative orphans move the line to reconciliation-required rather than being retried or assumed failed. You get the vendor, the operation and the token to check. ## 5. The seven channels — https://adbuymcp.com/channels One audience, compiled seven ways. Each channel below shows the vocabulary it actually accepts, the score AdBuyMCP gives that translation, and the limit it will not pretend past. The scores run from 85 to 40, and the 40 is on this page in the same size type as the 85. Every media platform can list its channels. What almost none of them will tell you is how much of your audience definition survives the trip into each one, because the answer is often "not much" and saying so costs a sale on the day. Publishing the number buys back more than it costs: if you have been told that a billboard can target IT decision makers you already know it is not true, and the vendor who says so first is the one they believe about the rest of the plan. ### What a fidelity score is A fidelity score is not a quality score, not a forecast and not a model output. It is a published, deterministic grade of how closely a channel can express an audience at all. A person grade means the vocabulary can carry a person-level claim; household, content, geo-cohort and venue each degrade one step further from the individual. Four of the seven are fixed constants — audio is always 65, addressable TV always 55, DOOH always 50, cinema always 40, however well your brief matched — because those vocabularies cannot express more than that. Three vary, across three discrete values each: CTV at 85, 65 or 60 depending on marketplace match strength, search at 82, 68 or 45 depending on whether you supplied intent, and social at 62, 48 or 38 depending on whether you supplied interests. One further rule applies everywhere: naming a city the UK geography table does not map costs 15 points on every channel, and the rationale names the city. ### Plan floors, enforced when the plan is built - Paid search: £100 - Digital out-of-home: £100 - CTV & streaming: £140 per week - Podcast & audio: £250 - Paid social: £250 - Cinema: £500 per week - Addressable & linear TV: £1,000 Enforced at plan generation: an under-floor line is dropped and its budget redistributed with an explicit note, and a manual plan is rejected against the same numbers. Vendor-side minimums are separate and bite later — Sky AdSmart's £3,000 burst is the one most likely to surprise a first plan. Q: Do I have to buy all seven channels? A: No, and at most budgets you should not. A broad seven-channel plan at £2,000 a month is operationally and statistically weak, and spreading a small budget across every channel is the fastest way to make none of them measurable. Two or three channels with enough weight to prove something is the shape that works. The plan generator funds the lines the budget can actually carry rather than putting a token line on every channel. Q: What is a lawful-basis manifest? A: A record, attached to every compiled channel plan, of which data made that targeting lawful: the provider of each segment, the basis it relies on — consent-chained, legitimate interest, or outside scope entirely — and the warnings raised while compiling it. It exists so that the answer to "where did this audience come from?" is a document rather than a conversation, and so that a DPO reviewing a campaign is reading the same record the compiler acted on. Q: Who decides which segments get used — the AI or a lookup table? A: The table. The language model extracts semantics from your brief: what kind of person, what interests, what geography, what intent. Deterministic registry code then matches those semantics against catalogue keywords and emits known segment identifiers. The model never freehands a vendor id, which is why a compiled plan can be audited line by line and why a hallucinated segment cannot reach a buy. Q: Are these the complete segment catalogues? A: No. The examples shown on each channel page are drawn from the sandbox-scale slices we ship with, which are real taxonomies at reduced size — the AdSmart attribute catalogue, the Vistar and Bombora taxonomies, an IAB Audience Taxonomy 1.1 slice, DCM packs, Acorn and Mosaic tables. The full exports load into the same shapes, so catalogue size changes neither the compiler behaviour nor the fidelity grade. No catalogue here has yet been loaded for a live campaign, because none has run. ### CTV & streaming TV — https://adbuymcp.com/channels/ctv-streaming The one channel where a job title survives the translation almost intact. CTV is where AdBuyMCP scores highest, because it is the only one of the seven channels where the buying vocabulary can carry a person-level claim honestly. A brief naming technology decision makers compiles into consent-chained marketplace segments — Bombora Company Surge topics, Dun & Bradstreet IT decision-maker segments — layered over city geography and a weekly frequency cap. Where at least one of those segments matches strongly, the compiled plan is graded person-level and scores 85. Where the matches are weak the grade drops to household and the plan says so, and where nothing matches it falls back to geo-city household targeting at 60 rather than pretending the segment was found. Fidelity: 60–85 out of 100, best grade person. It varies across 3 discrete bands. Fidelity bands: - 85 (person). Person/household-level via consent-chained marketplace segments (Bombora/D&B). When: At least one marketplace segment matches the persona strongly. - 65 (household). Marketplace segments matched at low confidence only; graded household-level pending review. When: Segments matched, but weakly. The plan is flagged for human review rather than promoted. - 60 (household). No marketplace segment matched; falling back to geo-city household targeting. When: Nothing in the catalogue matched the persona's terms. Targeting vocabulary this channel actually accepts: - Bombora Company Surge topics, from Bombora via DSP marketplace. Lawful basis: consent-chained. For example: DevOps · Cybersecurity · SaaS purchasing · AI & machine learning - Dun & Bradstreet firmographic segments, from D&B via DSP marketplace. Lawful basis: consent-chained. For example: IT decision makers · SMB owners - Experian audience segments, from Experian via DSP marketplace. Lawful basis: consent-chained. For example: Affluent professionals · Financial decision makers How the compiler works here: The persona's semantic core is matched against the marketplace catalogue by keyword, and the top four detailed matches become segment ids on the targeting spec. Match confidence is consumed by the fidelity grade rather than ignored: a person-level grade requires at least one strong match and never a lone low-confidence hit. Geography resolves to city and, where the persona names cities the UK geo table does not map, the score is penalised and the rationale says which ones. Age and gender bounds are carried only where the persona actually states them, because defaulting an absent bound to the full range turns "unspecified" into a positive claim you never made. Supply rails that reach it: StackAdapt, Amazon Ads (DSP sub-seat), Roku, Vibe (US). Creative requirements, enforced as schemas rather than advised: - VAST video: 16:9 ProRes 422 HQ mezzanine, 1920×1080, 25fps for the UK, ≥20Mbps. Platforms transcode down. - Storyboard: Generated with VAST metadata attached, so the creative and its trafficking record are one object. - Supers and loudness: Supers inside 90% title-safe; audio to EBU R128. Measurement: Delivery is provider-reported. Response is observed through your own analytics against the campaign's immutable UTM key. Causation needs a powered geo-lift test, and CTV can carry one where the buy is city-level rather than national. The exposure ledger — the thing that would let CTV deduplicate against DOOH and show a sequential journey — is designed, schema-backed and unit-tested, and is currently written to only by the sandbox: a live report proves aggregate delivery, not household exposure, so nothing mints household identities from live data. THE LIMIT, in this channel's own terms: Person-level here means the segment provider asserts a consent chain, not that AdBuyMCP has verified every link in it. The segments carry a data CPM uplift — roughly £2.50 per thousand on Bombora, £3.00 on D&B — which is real money and appears in the plan rather than being absorbed. And an 85 is still not a match on a named individual: it is the best available projection of one. Q: Can I target specific job titles on CTV? A: Not directly, and nobody can. What you can do is buy consent-chained B2B segments that are built from job-title and intent data — Bombora Company Surge, D&B IT decision makers — and have them delivered against streaming households. AdBuyMCP grades that person-level and scores it 85 out of 100 when the match is strong, which is the honest description of a very good proxy rather than of an exact match. Q: What happens if my audience matches nothing? A: The plan still compiles, at 60 out of 100, on geo-city household targeting, and the fidelity rationale says exactly that: no marketplace segment matched. We do not silently substitute a broader segment and present it as your audience. If 60 is not good enough for the money involved, that is a decision you get to make before you spend it. ### Podcast & digital audio — https://adbuymcp.com/channels/podcast-audio You are not buying a person here. You are buying what they listen to. Audio compiles to a content projection rather than a person: show and genre allowlists, publisher behavioural segments, and city geography derived from IP. AdBuyMCP scores it 65 out of 100 and grades it content-level, and the rationale is explicit that person-level is reached only where a publisher segment is consent-chained. One detail worth noticing: dayparts are set from the persona itself. A brief with firmographics — SIC codes or seniority — buys the morning and evening commutes, because that is when business audio is consumed. Everything else buys daytime and evening. Fidelity: 65 out of 100, best grade content. This is a fixed constant: the vocabulary cannot express more than this however well the brief matches. Fidelity bands: - 65 (content). Content + geo projection: show/genre allowlists and publisher segments proxy the persona; IP-derived city geo. Person-level only where publisher segments are consent-chained. When: Standard. Audio does not have a higher band, because the vocabulary does not support one. Targeting vocabulary this channel actually accepts: - Genre allowlists, from Spotify / Acast / DAX genre targeting. Lawful basis: legitimate-interest. For example: Technology · Business & finance · News & politics - Named-show packages, from DAX / Acast private marketplaces. Lawful basis: legitimate-interest. For example: UK tech flagship shows, bought as a PMP deal - DAX behavioural segments, from Global DAX. Lawful basis: consent-chained. For example: Business decision makers - Spotify topic targeting, from Spotify. Lawful basis: consent-chained. For example: Tech & gadget interest How the compiler works here: The persona is matched against the audio catalogue and the top five matches are split three ways by their identifier prefix: genres become an allowlist, named shows become a private-marketplace deal specification, and publisher segments stay as segments. Dayparts are chosen from whether the persona carries firmographics, which is the closest thing to a behavioural inference the compiler makes and it is a defensible one. Geography resolves to city. Supply rails that reach it: Spotify, StackAdapt, Amazon Ads (the DAX pipe), Global DAX, Acast. Creative requirements, enforced as schemas rather than advised: - 15 and 30 second scripts: Generated with voiceover direction and SFX notes, not just copy. A script with no read direction is half a brief. - Voiceover: UK male, female and neutral voices available as a separately charged asset at £1.25 per read. - Music bed: 5 to 180 seconds, generated, 50p at the floor. Measurement: Audio is a strong response channel and a weak causal one at small budgets: a vanity URL or promo code gives you observed response cheaply, and a powered lift test usually needs more spend than a podcast line carries on its own. Podscribe pixel and promo-code attribution is wired and client-built; no live campaign has produced rows through it yet. THE LIMIT, in this channel's own terms: Digital audio was around £410m in the UK in 2025 under the IAB's digital definition, growing about 14% — desk research from August 2026, not a product metric. It is a real channel but a small one, and a seven-channel plan that puts a token audio line in for completeness is worse than one that leaves it out. Sixty-five out of a hundred is a content projection: you are reaching people who listen to business podcasts in London, not the specific business decision makers on your account list. Q: Can I advertise on a specific podcast? A: Yes, through a private marketplace deal, and the compiler produces the named-show package as a PMP deal specification rather than as a targeting parameter. What it will not do is promise you a show it has no deal for. Named-show buying on DAX and Acast is an assisted rail on the supply page, which means the deal is negotiated rather than clicked. Q: Why does my audio plan buy the commute? A: Because your brief carried firmographics — an SIC code or a seniority level — and business audio is consumed on the way to and from work. If the brief has no firmographic signal the plan buys daytime and evening instead. It is a rule in the compiler rather than a model's guess, so you can disagree with it and override it. ### Addressable & linear TV — https://adbuymcp.com/channels/addressable-tv The fastest-growing channel on this list, and the one that needs the most paperwork. Addressable TV compiles to a household-level proxy: Experian Mosaic and CACI attribute combinations intersected with postcode districts and mapped to BARB television regions. AdBuyMCP scores it 55 out of 100. The rationale is precise about where the loss is — geo precision is strong, and firmographics are approximated by household affluence and occupation attributes — which is the honest description of buying "senior decision makers in business" as a Sky AdSmart attribute rather than as a verified job title. UK addressable TV was £1.84bn in 2025 and grew 37% on AA/WARC figures, which makes it the strongest growth engine in this channel set — though nothing here buys it yet. Fidelity: 55 out of 100, best grade household. This is a fixed constant: the vocabulary cannot express more than this however well the brief matches. Fidelity bands: - 55 (household). Household-level proxy: Experian Mosaic/CACI attribute combinations ∩ postcode districts. Strong geo precision; firmographics approximated by household affluence/occupation attributes. When: Standard, and the only band. A television set does not know who is in front of it. Targeting vocabulary this channel actually accepts: - Experian Mosaic groups, from Sky AdSmart / Experian. Lawful basis: legitimate-interest. For example: City Prosperity · Prestige Positions - Occupation attributes, from Sky AdSmart / Experian. Lawful basis: legitimate-interest. For example: Business owners & directors · Senior decision makers in business - Affluence and life-stage attributes, from Sky AdSmart / Experian. Lawful basis: legitimate-interest. For example: Household income £75k+ · Young professionals · Early technology adopters - Postcode districts and TV regions, from UK geography. Lawful basis: outside-scope. For example: EC1, EC2, E1, N1, SE1, SW1, W1, WC1 → London · M1–M15 → Granada How the compiler works here: The persona is matched against the AdSmart attribute catalogue and the top four attributes become the household selection. Cities resolve to their postcode districts and to their BARB macro region — London, Granada, Central, Yorkshire, West, STV — through an explicit geography table rather than a model's guess. Where the persona names a city with no UK mapping the score is penalised and the rationale names the city, so an unmapped geography is visible rather than silently dropped. Supply rails that reach it: Sky AdSmart — booking workflow, ITV Planet V — booking workflow, Comcast Universal Ads — typed stub, partner-gated, buys nothing today. Creative requirements, enforced as schemas rather than advised: - TV script: Generated with Clearcast substantiation notes attached, because a claim you cannot substantiate is a clearance rejection rather than a copy problem. - Mezzanine: 16:9 ProRes 422 HQ, 1920×1080, 25fps, ≥20Mbps, supers inside 90% title-safe, EBU R128. Measurement: Delivery comes from the sales house, through the reconcile step of the booking workflow rather than from an API — and no addressable-TV delivery statement has yet been ingested. Causation is possible in principle where a campaign is regional enough to leave a control group, which buying by postcode district and BARB region makes straightforward to arrange here. Household-keyed deduplication against CTV is designed and unit-tested but is exercised only by sandbox data today. THE LIMIT, in this channel's own terms: Two things, and the second is the bigger one. Fifty-five out of a hundred is the whole story on targeting: you are buying households whose Mosaic profile over-indexes on the attribute you asked for, not the people you asked for. And we do not buy UK broadcast inventory today at all. Universal Ads — the one integration that would cover ITV, Sky and Channel 4 in a single partnership — is a typed stub whose every live method fails closed, blocked on partner approval and a per-request signing key that does not exist. Sky AdSmart and ITV Planet V generate booking paperwork rather than making the booking. No addressable-TV delivery has ever been reconciled here. Anyone offering you one-click addressable TV in the UK is describing something else. Q: Is this the same as buying a TV spot? A: No. A spot buys everyone watching a programme. Addressable TV buys the households you selected, wherever they are watching, by swapping the ad in at the set-top box or the stream. That is why it is bought on household attributes and postcode districts rather than on programme and time, and why the fidelity question is about how well Mosaic groups proxy your audience rather than about how many people watched. Q: Why is 55 out of 100 worth buying at all? A: Because the alternative on television is worse, not better. A 55 that you can see is a more useful number than an unlabelled demographic buy, and addressable TV grew 37% to £1.84bn in 2025 on AA/WARC figures, which tells you the money is moving rather than that the trade is a good one. The point of publishing the score is that you get to decide whether it is worthwhile for your budget, rather than finding out afterwards. ### Cinema — https://adbuymcp.com/channels/cinema The lowest fidelity score on the site. Cinema compiles to a venue and film projection: DCM audience and genre packs, plus a venue list built from your persona's postcode districts. We grade it venue-level and score it 40 out of 100 — the lowest score in the system, published in the same size type as the 85 on CTV, because if you only see the good number you learn nothing about the method. Two honest notes. A cinema buy involves no personal data at all, so it sits outside UK GDPR scope, which is in your favour. And the Acorn catchment licence that would make the venue ranking real is not held, so the catchment over-index figures on a sandbox booking brief are fixtures rather than data. Fidelity: 40 out of 100, best grade venue. This is a fixed constant: the vocabulary cannot express more than this however well the brief matches. Fidelity bands: - 40 (venue). Venue/film projection: DCM audience packs + venues whose catchments over-index target Acorn/Mosaic types. No personal data — outside UK GDPR scope. When: Standard, and the only band. A cinema screen has no idea who bought the ticket. Targeting vocabulary this channel actually accepts: - DCM audience packs, from Digital Cinema Media. Lawful basis: outside-scope. For example: ABC1 Adults · Premium London - DCM genre packs, from Digital Cinema Media. Lawful basis: outside-scope. For example: Sci-fi & action · Drama & awards - Upmarket venue packs, from Pearl & Dean. Lawful basis: outside-scope. For example: Everyman and Curzon circuits - Catchment ranking, from CACI Acorn — licence pending, not in the product today. Lawful basis: outside-scope. For example: Intended: venues ranked by catchment over-index. Today: venue ids are built from the persona's postcode districts, and the over-index figures printed on a booking brief are deterministic sandbox fixtures. How the compiler works here: The persona is matched against the cinema pack catalogue, and the resulting packs are combined with a venue list derived from the persona's first eight postcode districts. Flights default to two weeks, because that is how cinema is sold. No personal data enters the specification at any point, which is why this is the one channel whose lawful-basis manifest is trivial. Supply rails that reach it: Digital Cinema Media, Pearl & Dean. Creative requirements, enforced as schemas rather than advised: - DCP: 2K flat, 1998×1080, 24fps, 5.1 audio, 60 seconds maximum. - Clearance: Routed via DCM Cinemapper with a CAA clearance certificate before print traffic. The workflow holds the state; it does not advance on a timer. - Storyboard: Generated for the cinema aspect and duration rather than reused from the CTV cut. Measurement: Cinema exposure is modelled at postcode-sector level, which in principle gives frequency deduplication against DOOH in the same catchment. What it cannot give you is a person-level exposure record, and no cinema seller in the UK can. Delivery arrives through the reconcile step of the booking workflow, and no cinema delivery statement has yet been ingested here. THE LIMIT, in this channel's own terms: Cinema was around £220m in the UK in 2025 on AA/WARC figures: small, and volatile. It is a distinctive attention product and a poor recurring-revenue wedge. We are emphatically not "the first self-serve cinema" — DCM has run a campaign planner and established buying workflows for years, and the claim is one we have explicitly retired. Cinema here is a booking-paperwork workflow with an eighteen-day lead time, no cinema delivery has ever been reconciled through it, and the Acorn licence behind the catchment ranking is not held. What is different is cinema inside a neutral cross-channel plan with a common audience definition and a common report. Q: Why publish a score of 40? A: Because it is the true one, and because a page that only publishes the 85 is a page that has told you nothing about how the other six were arrived at. A venue projection is genuinely worth about 40 out of 100 as a representation of a named audience. It can still be an excellent buy — a full screen with the lights down has an attention profile nothing else on this list matches — and those two facts are not in conflict. Q: Is cinema really outside GDPR? A: The targeting is. Nothing personal enters a cinema specification: it is audience packs, genre packs and venue catchments, all of them aggregate. That is why this channel's lawful-basis manifest is the shortest of the seven. Your own measurement of the campaign may involve personal data, and that is covered separately. ### Digital out-of-home — https://adbuymcp.com/channels/dooh A billboard cannot know who you are. It can know who walks past it. DOOH compiles to a property of a place rather than a property of a person: audience segments, OpenOOH venue types, postcode polygons and dayparts. AdBuyMCP grades it geo-cohort at 50 out of 100, and the compiler records that the delivery data is aggregated and therefore outside UK GDPR scope. One thing this page will not claim, because the code does not do it: there is no movement-indexed screen ranking in the product. The compiler emits an empty ranking list, and the over-index figures on hand-picked inventory are a constant. This is the channel where the gap between honest and dishonest targeting language is starkest — you are buying a rail station at commuter hours because business professionals are known to be there, not buying business professionals. Fidelity: 50 out of 100, best grade geo-cohort. This is a fixed constant: the vocabulary cannot express more than this however well the brief matches. Fidelity bands: - 50 (geo-cohort). Statistical screen property: screens ranked by persona over-index (movement data), venue types + geo polygons + dayparts. Aggregated — outside UK GDPR scope for delivery. When: Fixed. DOOH always grades 50 regardless of how well the persona matched, because the vocabulary supports no better claim. The ranking half of that rationale describes the intended design rather than shipped behaviour. Targeting vocabulary this channel actually accepts: - OpenOOH venue types, from OpenOOH. Lawful basis: outside-scope. For example: Rail stations · Underground · Office buildings & lifts · Roadside digital · Gyms · Malls - Vistar audience layer, from Vistar Media. Lawful basis: outside-scope. For example: Eyeota-D&B IT decision makers · Business professionals · Affluent shoppers - Adsquare ID-Precise segments, from Adsquare, consent-attested — contract pending, no data flowing. Lawful basis: consent-chained. For example: Tech-employer commuters, London · Frequent cinema-goers - Geo polygons and dayparts, from UK geography. Lawful basis: outside-scope. For example: Postcode-district polygons intersected with commute hours How the compiler works here: Venue types are selected by matching the persona's terms against the OpenOOH catalogue, and audience segment ids come from the Vistar and Adsquare layers. Geography becomes postcode polygons rather than a radius, and dayparts narrow the buy to the hours the cohort is present — the commute on a brief carrying firmographics, and not on one that does not. What the compiler does not produce is a ranked screen list: that field comes back empty, and screens are chosen by venue type, segment and polygon. Supply rails that reach it: Perion (Hivestack), StackAdapt, Vistar Media. Creative requirements, enforced as schemas rather than advised: - Portrait: 1080×1920, rendered to a real SVG at the exact screen resolution. - Landscape: 1920×1080. - Banner: 1400×400. - Copy limit: Headline capped at seven words, enforced in code rather than advised. A DOOH headline nobody can read at 4mph is not creative, it is spend. Measurement: Proof of play comes from the seller. Causation is geo-lift, and out-of-home is the channel that design suits best, because the buy is already made by place. The sequential journey — saw the billboard, then got the CTV ad — is a real feature of the exposure ledger and it currently runs on sandbox exposure data only: no inbound ingest route for viewshed panels exists yet, so treat the journey view as a working mechanism awaiting live input rather than as evidence about your campaign. THE LIMIT, in this channel's own terms: UK DOOH was around £955m in 2025 on IAB and AA/WARC figures, taking a growing share of out-of-home, though OOH overall was still contracting in the Q2 2026 IPA Bellwether. Fifty out of a hundred means the screen is a reasonable statistical bet, not that the person in front of it is your buyer. Two harder limits: there is no movement-index ranking in the product, and no DOOH rail can be driven end to end from a compiled persona in live mode — Hivestack activation needs hand-picked unit ids because venue and postcode semantics have no verified vendor-id translation, and Vistar has no live client at all. Blindspot and AdQuick both have better screen-level inventory tooling than this does today. Q: Are you tracking people's phones? A: No, and less than you might assume even in intent. The compiler records that DOOH delivery data is aggregate and outside UK GDPR scope, and where a consent-attested segment is referenced the lawful-basis manifest names the provider that attested it. Beyond that: the Adsquare movement-data contract does not exist yet, so no movement data reaches the product at all. Nothing in a DOOH plan identifies an individual. Q: Can I pick specific screens? A: Yes, and on this channel it is the honest route rather than a fallback. Picking billboards, shows, channels and venues by hand is a first-class path through the product. There is no screen ranking to overrule: the compiler returns an empty ranking list, because the movement data that would order screens by persona over-index does not reach the product at all. ### Paid search — https://adbuymcp.com/channels/paid-search One search plan. Two execution rails. Neither of them writes your keywords for you. Search is the channel where AdBuyMCP's fidelity score is most sensitive to the quality of your brief, and that is deliberate. A brief carrying buyer-authored intent topics compiles to an active-intent projection and scores 82. A brief with interests but no intent scores 68, and the rationale says plainly that interests are useful seeds for keyword planning but are not evidence of active product intent. A brief with only a persona label scores 45, with a rationale that tells you what is missing: product and offer context, before the keyword plan is activation-ready. Google Ads and Microsoft Advertising both execute the same neutral plan. Fidelity: 45–82 out of 100, best grade content. It varies across 3 discrete bands. Fidelity bands: - 82 (content). Active-intent projection from buyer-authored search topics, with native negative keywords and city geo. When: The brief names the search topics your buyers actually type. - 68 (content). Interest-to-query projection: useful seeds for keyword planning, but not evidence of active product intent. When: The brief names interests rather than intent. - 45 (content). Persona-label fallback only; product and offer context must be supplied before the keyword plan is activation-ready. When: The brief is a persona label and nothing else. Targeting vocabulary this channel actually accepts: - Buyer-authored intent topics, from Your brief. Lawful basis: outside-scope. For example: Plain-language seeds only. The platform adapter resolves keyword ideas and match data. - Native negative keywords, from Compiled from the brief's exclusions. Lawful basis: outside-scope. - City geography, from UK geo table. Lawful basis: outside-scope. How the compiler works here: Search has no segment catalogue behind it, unlike the five channels above: it compiles from your own words with no registry lookup at all. The persona's intent topics become plain-language seeds; the provider's own keyword planner resolves them into ideas with volume and bid data. Nothing auto-buys. The buyer explicitly applies buy words and negatives to a version-fenced draft, and provider suggestions never become live keywords on their own. Positive and negative terms that collide once normalised for Unicode, whitespace and case are rejected atomically rather than partially applied, and an absent Microsoft bid or volume value is treated as unavailable rather than as zero. Supply rails that reach it: Google Ads, Microsoft Advertising. Creative requirements, enforced as schemas rather than advised: - Responsive search ads: 3 to 15 headlines at 30 characters each, 2 to 4 descriptions at 90 characters. The limits are schema constraints, so over-long copy fails at the edge rather than being silently truncated on the way to the provider. Measurement: Search is the easiest channel to measure and the hardest to claim credit on, which is why it sits inside a cross-channel plan here rather than being sold as the whole answer. Delivery and click data come from the provider. Response is observed through your analytics on the campaign's immutable UTM key. Initial canaries are non-conversion only — clicks, not conversions — until an approved conversion definition is bound to an exact provider-owned goal, read back before staging and reverified at resume. THE LIMIT, in this channel's own terms: Search was £17.9bn of a £46.7bn UK market in 2025, on AA/WARC and IAB UK figures. You are almost certainly already buying it well, and we are not going to out-optimise your existing search team. What it does is stop search being planned in a different room from the £3.4bn of addressable TV, DOOH, audio and cinema around it, and it deduplicates the audience across the two. If all you want is better search management, buy something else. Q: Does the AI write my keywords? A: No, and the refusal is structural rather than a setting. The model turns your brief into plain-language seeds. The provider's keyword planner resolves those into ideas with real volume and bid data. You then explicitly apply buy words and negatives to a version-fenced draft. Provider suggestions never auto-buy at any point in that chain. Q: Why did my search plan only score 45? A: Because the brief was a persona label with no product or offer context, and a keyword plan built from a persona label alone is not ready to activate. The rationale on the plan says exactly that. Add what you sell and what the offer is, and the same brief compiles at 68 or 82. ### Paid social — https://adbuymcp.com/channels/paid-social Vendor-neutral until the moment of execution, then exactly as native as it needs to be. Paid social compiles to a platform-owned interest and demographic projection using your own terms, and scores 62 out of 100 when interests are supplied. Without interests it drops to 48 on demographics alone, and to 38 on a bare city buy — where the rationale tells you to add buyer-authored interests before treating the result as a faithful projection of your persona. Meta and TikTok execute the same neutral audience semantics, and the exact audience availability is resolved per platform rather than promised in advance. Fidelity: 38–62 out of 100, best grade content. It varies across 3 discrete bands. Fidelity bands: - 62 (content). Platform-owned interest and demographic projection using buyer-authored terms; exact audience availability is resolved per platform. When: The brief supplies interest terms. - 48 (content). Broad city and demographic projection only; no interest signal was supplied. When: Demographics but no interests. - 38 (content). Broad city targeting only; add buyer-authored interests before treating this as a faithful persona projection. When: Neither interests nor demographics. Targeting vocabulary this channel actually accepts: - Buyer-authored interest terms, from Your brief, resolved per platform. Lawful basis: consent-chained. - IAB Audience Taxonomy 1.1, from IAB. Lawful basis: consent-chained. For example: Technology & computing · Business & industrial · Personal finance - Platform demographics, from Meta / TikTok. Lawful basis: consent-chained. How the compiler works here: Interest terms are taken from the brief verbatim — normalised and de-duplicated, never enriched — and handed to each platform's own audience resolution. The IAB Audience Taxonomy in the table below is registered for this channel, but the social compiler passes no catalogue entries at all, so nothing is matched against it. What it never emits is a platform audience identifier: the interests travel as plain-language hints rather than as Meta or TikTok audience ids, and exclusions are not compiled into audience ids either. That is the point and the limit of vendor-neutral semantics: AdBuyMCP holds one definition of the audience and each platform resolves it into whatever it offers, rather than you maintaining two divergent definitions in two ad managers. Supply rails that reach it: Meta Ads, TikTok Ads. Creative requirements, enforced as schemas rather than advised: - Bounded social copy: Headline 40 characters, description 30. Conservative portable limits that include TikTok's tighter 100-character text envelope, so one creative is valid on both platforms. - Video provenance: AI-generated video submitted to TikTok carries a SELF_DISCLOSURE declaration. Missing provenance blocks before any provider call, and the declaration must survive the provider readback at resume. Measurement: Same as search: provider delivery, observed response through your own analytics on the immutable campaign key, and no live conversion optimisation until the conversion definition is bound to an exact provider-owned action. Initial canaries are traffic and reach only. THE LIMIT, in this channel's own terms: Social was £11.5bn of the UK market in 2025 on AA/WARC figures, and you are already buying it. Meta and TikTok optimise their own inventory better than any third party will, and nothing here changes that. Sixty-two out of a hundred is an interest projection. The reason paid social is here is not that we buy it better — it is that a plan which ignores the channel carrying most of your budget cannot honestly claim to have deduplicated your audience or measured incremental reach. Q: Why would I buy social here instead of in Ads Manager? A: For most buyers, on its own, you would not, and the page says so. The reason social is in the product is deduplication and incrementality: without the channel carrying most of your budget, a cross-channel frequency number is wrong and an incremental-reach claim is unsupportable. If you are only buying social, Ads Manager is free and better. Q: What does vendor-neutral actually mean here? A: That the audience is defined once, in AdBuyMCP's own semantics, and each platform resolves that definition into whatever it offers. You are not maintaining one audience in Meta and a differently-worded one in TikTok and hoping they mean the same thing. What it does not mean is that the two will reach identical people — exact availability is resolved per platform, and the plan says so. ## 6. Supply status — https://adbuymcp.com/supply 25 registered connectors, all functional in a deterministic sandbox. Live evidence is graded on a six-level ledger, and no rail has yet spent money or reconciled a real report. Most platforms answer "what can you actually buy?" with a wall of logos, which is unfalsifiable and therefore worth nothing to the person asking. This is the falsifiable version, and the grades are not ours to soften: they come from our own evidence ledger, which states that a mocked HTTP test proves a client was written, not that a vendor accepted the request, spent money, or returned real delivery. Two of the six levels have never been reached by anything, and a continuous-integration check enforces that those two counts stay honest. One connector — Vibe, on 13 August 2026 — has had read-only calls accepted by a live vendor API: OAuth, advertiser discovery and a 549-unit catalogue read. That is the entirety of the live vendor evidence. No connector has ever executed a buy. ### The six-level evidence ladder - Level 1 of 6, Sandbox only — 0 of 25 connectors. Deterministic behaviour for planning and rehearsal. No live path is claimed at all. - Level 2 of 6, Stub — 6 of 25 connectors. The adapter models the vendor's shapes and drives the sandbox, but every live method fails closed and makes no vendor call whatsoever. Blocked on a contract, a seat or a published schema. - Level 3 of 6, Client built — 18 of 25 connectors. A live HTTP client exists and passes mocked tests. In our own words, that proves the client is written — not that a vendor accepted the request, spent money, or returned real delivery. - Level 4 of 6, Contract tested — 1 of 25 connectors. A real vendor API accepted real calls from this client. It does not mean a buy has been executed. - Level 5 of 6, Spend tested — 0 of 25 connectors. A controlled paid activation has run through this rail. Nothing has reached this level, and a continuous-integration check keeps the count at zero until something genuinely earns it. - Level 6 of 6, Report verified — 0 of 25 connectors. Provider report rows have been reconciled to an invoice through this rail. Nothing has reached this level, and a continuous-integration check keeps the count at zero until something genuinely earns it. ### Every connector, with its blocker #### Buying API (12 of 25) - StackAdapt [stackadapt] — Client built, priority P0. Programmatic CTV, audio and DOOH through one GraphQL seat. The broadest single rail here. Channels: CTV & streaming, Podcast & audio, DOOH. Blocker: A contracted seat, then a controlled paid activation reconciled to a StackAdapt invoice. Access, which is not a signup: Live inventory browse is not available: the client deliberately reports that StackAdapt has no public inventory-browse API, so catalogues shown in the sandbox are sandbox catalogues. - Google Ads [google-ads] — Client built, priority P0. Paid search, with buyer-applied keywords and negatives on a version-fenced draft. Provider suggestions never auto-buy. Channels: Paid search. Blocker: The design-partner canary, passed for this rail specifically: staged PAUSED, human-recorded creative decision, separately granted resume, capped canary, then spend reconciled to the Google account total. Access, which is not a signup: Initial canaries are clicks-only. Live conversion optimisation fails closed until an approved conversion definition is bound to an exact provider-owned goal, read back before staging and reverified at resume. - Microsoft Advertising [microsoft-ads] — Client built, priority P0. Paid search on Bing, planned against the same neutral brief as Google rather than as a separate campaign. Channels: Paid search. Blocker: The same canary gate, passed separately. A Google pass proves nothing about Microsoft. - Meta Ads [meta-ads] — Client built, priority P0. Paid social on Facebook and Instagram, staged PAUSED until a human approves the exact creative bytes. Channels: Paid social. Blocker: The canary gate, passed separately, including an ad-account and Page identity check. - TikTok for Business [tiktok-ads] — Client built, priority P0. Paid social on TikTok. AI-generated video carries a self-disclosure provenance declaration that must survive the provider readback. Channels: Paid social. Blocker: The canary gate, plus exact AI-generation provenance on every selected video. - Amazon Ads [amazon-dsp] — Client built, priority P0. Amazon DSP sub-seat for CTV and audio, with refresh-token rotation held under a single-writer lock. Channels: CTV & streaming, Podcast & audio. Blocker: A live test buy and a reconciled report. Access, which is not a signup: Requires a DSP sub-seat obtained through an authorised reseller. Not an open developer signup. - Spotify [spotify-ads] — Client built, priority P0. Spotify Ads API for podcast and music audio. Channels: Podcast & audio. Blocker: A successful write. A live probe on 13 August 2026 established only that the snake_case routes answer 400 or 403 — that is route existence, not an accepted request. Access, which is not a signup: Needs an Ads-API-enabled app, API-terms registration and per-ad-account user OAuth, which grants access to your own account rather than multi-tenant inventory. Podcast targeting remains topic and genre based in closed-test campaign APIs. - Roku [roku] — Client built, priority P0. Roku CTV. Reporting only. Channels: CTV & streaming. Blocker: Roku documents campaign and ad-group writes and creative operations as future functionality, so activation, pause, creative submission, forecast and audience sync all fail closed before any request. Only report pulls are implemented. Access, which is not a signup: Self-service Ads Manager accounts, not credential-free multi-tenant supply. - Perion (Hivestack) [hivestack] — Client built, priority P0. Programmatic DOOH. Channels: DOOH. Blocker: Live activation requires hand-picked Hivestack unit ids: venue and postcode semantics have no verified vendor-id translation, and live creative submission and report pulls both fail closed. A compiled DOOH persona cannot drive this rail end to end today. Access, which is not a signup: Keys come through a commercial relationship rather than open self-serve signup. - Vibe [vibe] — Contract tested, priority P1. Performance CTV in the United States. The only rail with any live vendor evidence at all. Channels: CTV & streaming. Blocker: Contract-tested covers the header and the inventory read only — OAuth, advertiser discovery and a 549-unit catalogue read, on 13 August 2026. Activation, pause, creative and reporting are all still client-built, forecast is sandbox, and the report download payload is undocumented. Its own gate reads: execute one controlled test buy and verify the asynchronous report payload end to end. Access, which is not a signup: US inventory in USD only. The adapter refuses a non-USD budget outright, and this rail is deferred from the UK queue until that changes. The 549 units are US units and are not UK reach. - Vistar Media [vistar] — Stub, priority P0. DOOH and cinema supply. Sandbox planning data only. Channels: DOOH, Cinema. Blocker: There is no live client. Forecast, activate, pause, creative submission, reporting and inventory listing all throw before any request. Vistar publishes no stable normalised reporting schema, and creative submission needs an advertiser domain, IAB categories, publisher scope and OpenRTB markup the current shape cannot carry. Access, which is not a signup: An approved trafficking seat plus separate bidder approval. Partner access, not developer signup. - Comcast Universal Ads [universal-ads] — Stub, priority P0. A typed client stub for the Universal Ads Marketing API. Channels: Addressable & linear TV. Blocker: Partner approval and a per-request RSA signing key, neither of which exists. Every live method fails closed. We do not buy ITV, Sky or Channel 4 inventory today, and this rail should not be read as implying otherwise. Access, which is not a signup: Developer registration is a form; the partner approval behind it is a commercial relationship, and we have not entered into one. #### Booking workflow (6 of 25) - Digital Cinema Media [dcm-cinema] — Client built, priority P0. Generates the UK cinema booking brief with audience and genre packs, drives the clearance step, and holds the IO state machine. Channels: Cinema. Blocker: Ingest and reconcile the first authoritative sales-house delivery statement. These rails automate the paperwork; they do not make the booking and they do not report the delivery. In live mode a report pull throws rather than fabricating rows, and a pause requires sales-house acknowledgement. - Pearl & Dean [pearl-dean] — Client built, priority P1. The same booking automation for the cinema circuits DCM does not carry. Channels: Cinema. Blocker: Ingest and reconcile the first authoritative sales-house delivery statement. - Sky Media [sky-adsmart] — Client built, priority P1. Sky AdSmart addressable TV, booked as a generated brief on household attribute combinations rather than bid on. Channels: Addressable & linear TV. Blocker: Ingest and reconcile the first authoritative sales-house delivery statement. - ITV [itv-planetv] — Client built, priority P1. ITV Planet V booking paperwork. Channels: Addressable & linear TV. Blocker: Ingest and reconcile the first authoritative sales-house delivery statement. - Global DAX [dax-deals] — Client built, priority P1. Deal paperwork for Global's digital audio supply. Channels: Podcast & audio. Blocker: Ingest and reconcile the first authoritative sales-house delivery statement. - Acast [acast] — Client built, priority P1. Podcast private-marketplace and named-show package paperwork. Channels: Podcast & audio. Blocker: Ingest and reconcile the first authoritative sales-house delivery statement. #### Audience data (3 of 25) - European Unified ID [euid] — Stub, priority P0. A consent-chained identifier for addressing first-party lists without a cookie. Channels: CTV & streaming. Blocker: An operator agreement. The adapter makes no vendor call, and the CTV compiler emits an empty EUID audience list — no compiled plan carries an EUID today. - CACI [caci-acorn] — Stub, priority P0. Acorn classifications, intended for ranking cinema and DOOH catchments. Channels: Cinema, DOOH. Blocker: A CACI data licence. No Acorn data is in the product: the stub's scores are deterministic pseudo-random output, and cinema venue ids are constructed from postcode districts rather than from a licensed catchment dataset. - Adsquare [adsquare] — Stub, priority P1. Consent-attested movement data, intended for DOOH screen ranking. Channels: DOOH. Blocker: An Adsquare contract. There is no movement index in the product: the DOOH compiler emits an empty screen-ranking list, and hand-picked inventory carries a constant over-index of 1.0. #### Measurement (4 of 25) - Plausible Analytics [plausible] — Client built, priority P0. Your own site's analytics, read through your token, so brand response is measured on data you already own and can audit. Channels: All seven. Blocker: A human-verified analytics dossier bound to the site and revision, and a known goal and revenue total reconciled, before the source counts as observed. - Google Analytics 4 [ga4] — Client built, priority P0. GA4 as an alternative brand-response source where that is what the advertiser already runs. Channels: All seven. Blocker: The same dossier and reconciliation gate as Plausible. - Podscribe [podscribe] — Client built, priority P0. Podcast attribution: pixel and promo-code response against audio delivery. Channels: Podcast & audio, CTV & streaming. Blocker: A live campaign with rows to reconcile. - Cint [cint] — Stub, priority P0. Brand-lift survey panels, intended for the fourth measurement verb. Channels: CTV & streaming, Podcast & audio, Addressable & linear TV, Cinema, DOOH. Blocker: A Cint contract, and a commissioned study. The adapter makes no vendor call and the measurement payload returns brand lift as an explicit placeholder. ### What has to happen before a rail is called live 1. Contract and credentialed probe. Hold the supplier contract and account scope, save the encrypted credentials, and run a probe against that approved account. Record the accepted request and response shapes, permissions, currency, account identity and idempotency behaviour in a supplier dossier bound to that connection revision. 2. Stage everything paused. The first outbound activation creates the campaign, ad group and ad PAUSED — DISABLED on TikTok — persists every provider id in an exact staged-resource manifest, and begins no spend at all. 3. A human approves the exact bytes. A named person records the provider's decision for the exact creative bytes, the immutable destination and the external reference. Approval records evidence; it cannot enable spend. 4. A separate grant enables it. Resume is a separately granted action that first proves the upstream hierarchy still equals the persisted manifest — no extra, missing or substituted child — and that the upstream account equals both the current credential account and the staged account. A mismatch fails before any write. 5. A tiny capped canary, then a rehearsed failure. Run a small canary against a capped commitment, confirm delivery in the provider's own UI, invoke the durable pause, and reconcile both a known-success path and an ambiguous one. An ambiguous outcome is never blindly retried; it goes to a human. 6. Reconcile to the invoice. Pull a non-empty authoritative report, preserve native dimensions and currency, and reconcile spend to the provider invoice or account total, with the exact export bytes uploaded by a signed human. Only then may a rail move to spend-tested and report-verified. Q: Can I buy media through AdBuyMCP today? A: Not self-serve. You can run the entire loop today in a sandbox that needs no credentials and no card — brief, persona, seven compiled plans, creative, approval, launch, delivery, measurement — and everything it returns is deterministic and repeatable. What is gated is real spend: no rail has reached spend-tested or report-verified on our own evidence ledger, so live buying happens inside a design-partner engagement rather than behind a signup form. Q: What does client-built actually mean? A: That a live HTTP client is written and passes tests against mocked responses. Our ledger is deliberately blunt about the limit of that: a mocked HTTP test proves the client is built, not that a vendor accepted the request, spent money, or returned real delivery. Eighteen of the twenty-five connectors are at this grade. It is real engineering and it is not evidence of supply. Q: Which rails have had a real vendor accept a real call? A: One. Vibe, on 13 August 2026 — OAuth, advertiser discovery and a 549-unit catalogue read, all read-only, all US inventory. Its own missing gate reads: execute one controlled test buy and verify the asynchronous report payload end to end. Within that same connector, activation, pause, creative and reporting are still client-built. Nothing here has ever executed a buy. Q: Why are cinema and addressable TV described as workflows rather than APIs? A: Because that is what they are. Digital Cinema Media and Sky AdSmart do not offer self-serve buying APIs. AdBuyMCP automates the paperwork: it generates the booking brief with the right audience packs, drives the clearance step and holds the IO state machine. We do not make the booking and we do not report the delivery — in live mode a report pull throws rather than fabricating rows, and a pause requires sales-house acknowledgement. No cinema or addressable-TV delivery has ever been reconciled here. Q: Your README lists Vistar as having a live client. This page says it is a stub. A: The code is right and the README is stale. Every one of Vistar's live methods — forecast, activate, pause, creative submission, reporting and inventory listing — throws before making any request, and the provenance ledger grades it stub. Where prose and code disagree anywhere on this site, the code wins. Q: Do you buy ITV, Sky or Channel 4? A: No. The Universal Ads integration, which would cover that inventory in one partnership, is a typed client stub whose every live method fails closed, blocked on partner approval and a per-request signing key that does not exist. Sky AdSmart and ITV Planet V are booking-paperwork rails rather than buying APIs. Any impression that we buy UK broadcast inventory today would be wrong. Q: What happens if a supplier's authority expires mid-campaign? A: Spend stops. An expired or blocked supplier dossier raises a spend lock before any scheduler work runs, sends one incident-bound emergency pause, and refuses to retry an ambiguous response. Only a signed human reconciliation clears it, and re-verifying then revoking again uses a fresh incident id rather than reopening the resolved one. ## 7. Measurement — https://adbuymcp.com/measurement Four verbs — delivered, responded, caused, remembered — kept visibly apart, because the entire value of a measurement claim is in which of the four it belongs to. Three are built, the fourth is a placeholder, and no campaign has yet produced a causal result because nothing has run live. What this page sells is the refusal, not a lift number. ### The four verbs #### Delivered — "Did the media run?" [shipped] What backs it: Normalised delivery rows from each provider, with the additive provider event counts kept side by side rather than merged. Reach and frequency are modelled against the persona universe using Sainsbury deduplication, and anything modelled is labelled modelled. How far the claim goes: Ratios and cost averages stay dated observations and are never summed across days. A blended CPM computed by adding two days of averages is a made-up number, and the payload refuses to produce one. #### Responded — "Did anyone do something afterwards?" [shipped] What backs it: Outcomes observed on your own property — pageviews, conversions and their value — through Plausible or GA4, matched on the campaign's immutable UTM key, plus promo codes, vanity URLs and pixel events. Podscribe covers audio attribution. How far the claim goes: This is correlation and the payload says so in its own note. Two exclusions are enforced rather than advised: the synthetic geo-panel baseline series never inflates response counts, and nothing dated before the flight start can be counted as a response to it. #### Caused — "Would it have happened anyway?" [shipped] What backs it: A matched-market geo-lift test over UK conurbations with a difference-in-differences readout, run from one button. Before it runs, a two-sample power calculation on daily observations decides whether the test can answer the question at all, at α = 0.05 and 80% power. How far the claim goes: Two limits, and the second is absolute today. If the design is underpowered the engine returns an underpowered verdict and the required sample days rather than a lift number — that refusal is the point of it. And a result is only graded causal when every geo outcome is observed, every delivery row maps to a live line, and control exposure is verified. Sandbox delivery rows are labelled modelled by construction, so no sandbox campaign can reach a causal grade, and none has: the seeded demonstration asserts its own claim as a modelled demonstration with an injected lift, and is not evidence about anything. #### Remembered — "Did it change what people think?" [placeholder] What backs it: Brand-lift study through Cint: an exposed-versus-control panel measuring recall and consideration. It is the P0 measurement rail for the fourth verb. How far the claim goes: Nothing here is a substitute for the other three. Brand lift tells you about attitude, not about revenue. Why it is published anyway: Not built. The API returns this verb with the status "placeholder" and the note that the study has not been commissioned in this build, and the Cint adapter is explicitly non-transacting pending a contract. It is on this page because a four-verb model that quietly ships three verbs is exactly the kind of rounding we exist not to do. ### Provenance labels, which keep the verbs apart - Provider-reported: The seller's own delivery numbers, as they were returned, in their native dimensions and currency. - Observed: Something that actually happened on a property you own and can audit, matched to the campaign by an immutable key. Correlated with the media, not attributed to it. - Modelled: An estimate produced by a formula rather than a count — deduplicated reach and frequency being the common case. Always carries the word. - Powered causal: The only label that supports the word "caused", and it is only issued by a geo-lift test that passed its own power analysis before it ran. A delivery row is graded observed only when the exact persisted campaign line it belongs to executed live. Everything else is modelled, including every row the sandbox produces, and a mixed set is reported as mixed rather than rounded up. ### One ledger, four keys, and no identity joins inside it Exposures across all seven channels write into one ledger, keyed on whichever of four identifiers the channel can supply. That single ledger is the mechanism behind cross-channel frequency deduplication and sequential journeys — and it is worth being exact about its status, because the mechanism and the data are different things. STATUS, stated exactly: The four-key spine is designed, schema-backed and unit-tested, and today it is fed only by the deterministic sandbox. There is no inbound ingest route for viewshed panels, CTV logs or IP logs, and no shipped code path writes a hashed advertising id, a EUID or a postcode sector from live data — only a household key from the sandbox graph. The refusal is deliberate: a live report proves aggregate delivery, not household exposure, so it must never mint identities that could later cross a vendor boundary. Treat the journey explorer as a working mechanism awaiting live input. Keys, in precedence order: - householdKey: An opaque household key issued by the key-translation service. The ledger never resolves it back to anything. - maidHash: A hashed mobile advertising id. Held for at most 90 days, then persistently stripped from the event, which re-grades the event rather than deleting it. - euid: The consent-chained European Unified ID, where one exists. - postcodeSector: The coarsest key and the one that always works. Cinema and DOOH usually land here. Why four and not one: Multi-key by design, because MAIDs are a declining asset. A spine that only worked on mobile advertising ids would degrade into uselessness on a schedule set by Apple and Google; this one loses precision a step at a time and says which step it is on. Where identity is allowed to happen: The ledger itself never performs an identity join. Resolution between key spaces happens in one place — the key-translation service — which is the only module we have that touches identity, and it is the module the 90-day rule is enforced in. Keeping that boundary in one file is what makes it auditable. ### Prove It One button that runs a matched-market geo-lift test, and refuses to run it when the answer would not mean anything. 1. Match the markets. Candidate UK conurbations are paired on their pre-campaign outcome series so that exposed and held-out markets were behaving the same way before the media started. 2. Run the power analysis first. A two-sample calculation on daily observations, defaulting to α = 0.05 and 80% power, returns either a powered verdict with the minimum detectable lift, or an underpowered verdict with the number of sample days the design would actually need. 3. Refuse, or read out. An underpowered design returns no lift figure at all. A powered one runs difference-in-differences against the held-out markets and reports the lift with its confidence interval. The refusal is the feature. An incrementality number from an underpowered test is not a weak result, it is a random one, and the reason this engine will not produce one is that a random number presented in a board pack is indistinguishable from a real one. The two refusals, as the buyer sees them: - Underpowered: "The test is not adequately powered: N days of post-exposure data are needed to detect an X% lift on this market's baseline variance, and this campaign has fewer. Reporting a number from it would be reporting noise." Returned with the required sample days and the minimum detectable lift, so you know what would fix it. - No holdout: "Every candidate control market received delivery from this campaign, so none is a holdout. Exclude those markets from targeting for the test window, or nominate a control that this campaign does not buy." A contaminated control is the most common way a lift test quietly measures nothing. It is caught before the readout rather than after it. ### Eligibility, asked before the money moves Whether your campaign can support a causal claim is a property of your spend, your geography and your conversion volume, not of ours, and it is worth knowing before you fund the campaign rather than after it. So all four verbs are graded at planning time rather than at reporting time. Delivered comes back as available once delivery starts. Responded comes back as needing a brief, needing instrumentation, not requested, or collecting. Caused comes back as eligible or not, with the reason. Where the design cannot be powered, the plan offers response and reach evidence instead and says that is what it is offering — which is a considerably less pleasant conversation to have at the point of sale than at the point of reporting, and a considerably more useful one. Q: Why publish a verb you have not built? A: Because the alternative is to describe a three-verb model and let the fourth appear later as a launch. The four verbs are the measurement model; brand lift is the one that answers whether anything changed in people's heads, and leaving it out would make the model look complete when it is not. The API returns "remembered" with the status placeholder and a note that the study has not been commissioned, and the Cint adapter is non-transacting pending a contract. That is what this page says too. Q: What does it mean that the geo-lift engine refuses tests? A: That if your spend, geography and conversion volume cannot detect a lift of a plausible size at 80% power, the engine returns an underpowered verdict and the number of days the design would need — and no lift figure. Most incrementality tools will happily return a number in that situation. The number is noise, and the harm is that it is indistinguishable from a real one once it is in a slide. Q: How long do you keep mobile advertising ids? A: At most 90 days. After that the hashed id is persistently stripped from the exposure event, which re-grades the event to a coarser key rather than deleting the record. The rule lives in the key-translation service, which is the only module we have that performs identity joins at all. Q: Can you attribute a conversion to a billboard? A: You can see a sequence, and that is a different claim. All seven channels write into one ledger, so the mechanism can show that a postcode sector exposed to a DOOH panel was later exposed to a CTV ad and later produced a conversion. Be exact about what feeds it: the ledger is fed only by the deterministic sandbox today, there is no inbound ingest route for viewshed panels, CTV logs or IP logs, and sandbox rows are labelled modelled rather than observed. So it is a working mechanism awaiting live input rather than a sequence anyone has yet seen in real delivery. Turning a sequence into a causal statement takes a powered geo-lift test in any case, and out-of-home is a good fit for one because the buy is bought by place. Q: Have you produced a causal result for anyone yet? A: No. Nothing has run live, and a geo-lift is graded causal only when every geo outcome is observed, every delivery row maps to a live line and control exposure is verified — conditions sandbox data cannot meet by construction. Any lift figure you see in a demonstration is injected sandbox data labelled as a modelled demonstration. The engine, the power gate and the refusal are real and tested; the result is not yet evidence about a real campaign, and we do not print one as if it were. Q: Do you use marketing-mix modelling? A: There is a modelled layer, and it is labelled modelled wherever it appears — deduplicated reach and frequency being the main case. What the product will not do is present a model output as evidence of causation. Modelled estimates and powered causal results are two of the four provenance labels precisely so that they cannot be read as the same thing. ## 8. The MCP server — https://adbuymcp.com/mcp AdBuyMCP ships a Model Context Protocol server: 76 tools covering the whole loop, from extracting a persona out of a sentence to reconciling a statement. Claude drives the loop. You keep the money and the approvals, and today that loop runs against the sandbox, because no rail is spend-verified. The MCP server is not a chat wrapper on a dashboard. Each of its 76 tools is a thin wrapper over exactly one REST endpoint — the same endpoint the browser calls — so the agent path and the human path share one tenant boundary, one wallet gate, one approval record and one audit trail. There is no second execution path with weaker rules, because there is no second execution path at all. The corollary is the more important half: 11 consequential operations are deliberately absent from the tool surface entirely. An agent prepares and proposes. A named human authorises money. AVAILABILITY, before the setup instructions: MCP access today is a private, single-tenant, self-hosted deployment behind a bearer token or your own tunnel. There is no public connector to add and no multi-tenant sign-in. Establishing real OAuth 2.1 for public multi-tenant MCP is an open launch gate: the server publishes protected-resource metadata and scope challenges and refuses to mix OAuth with static tokens, but it does not embed an authorisation server, and a static bearer token is suitable only for a single-tenant deployment. ### The 76 tools, grouped by job - Missions and authority (19). The control plane an agent works inside: a durable mission, a bounded mandate, ordered steps, and an exception that stops the run rather than being retried. list_agent_actions, prepare_agent_action, create_agent_mission, list_agent_missions, get_agent_mission, start_agent_run, list_agent_runs, get_agent_run, add_agent_step, complete_agent_step, request_agent_approval, advance_agent_run, raise_agent_exception, list_agent_exceptions, resolve_agent_exception, resume_agent_run, record_agent_decision, record_agent_evidence, finish_agent_run - Audience and planning (13). Turn a sentence into a persona, compile it to seven channels, and produce a funded plan — or pick the inventory by hand. list_personas, extract_persona, compile_persona, browse_inventory, generate_plan, create_manual_plan, list_plans, get_plan, update_plan_allocations, approve_plan, suggest_search_keywords, get_plan_search_keywords, set_plan_search_keywords - Creative (8). Generate channel-correct creative, precheck it against the format and clearance rules, revise it in your own words, and submit it. generate_creative, precheck_creative, revise_creative, set_campaign_creatives, submit_creative, get_asset_pricing, produce_asset, get_asset_status - Launch and lifecycle (9). Fund, launch, pause and resume, with readiness and the provider-attempt ledger visible rather than inferred. launch_campaign, pause_campaign, resume_campaign, list_campaigns, get_campaign, get_campaign_readiness, list_live_execution_grants, list_campaign_attempts, get_campaign_dead_letters - Delivery and measurement (13). The four verbs, the journey view, the audience sync, and the one button that refuses to run an underpowered test. get_delivery, get_measurement, get_campaign_tracking, set_campaign_tracking, import_outcomes, get_outcome_health, get_campaign_report, prove_it, get_campaign_insights, get_journeys, sync_audience, mmm_readiness, run_micro_mmm - Optimisation (4). Explainable proposals that wait for a human. An optimiser that moves budget on its own is a different product with a different risk profile. generate_optimizations, list_optimizations, approve_optimization, reject_optimization - Money, connections and audit (10). The wallet, the fee statement, the supplier dossiers, and the trail that says who did what. get_workspace_overview, get_brand_analytics, list_connections, get_supplier_dossier, set_connection, get_billing_summary, get_billing_statement, get_topup_status, top_up_wallet, get_audit ### The 11 operations an agent cannot reach at all These have no MCP tool. They are human-only signed REST actions, and an integration test proves that an agent attempting to approve its own work receives a 403 requiring human approval. - Resolving an approval - Resolving a campaign provider attempt - Acknowledging a booking-workflow step - Resolving an asset job - Replacing a mandate - Revoking a mandate - Closing a mission - Saving a supplier dossier - Verifying a supplier dossier - Issuing a live execution grant - Revoking a live execution grant ### What the agent may do, and where it has to stop Give AdBuyMCP a mission and bounded authority. It translates the audience, prepares the plan and the creative, surfaces the evidence and the exceptions, asks for an exact approval where its authority ends, executes through the supplier rails, and reconciles what came back. No rail is spend-verified, so today that execution runs in the sandbox. - Mission: A durable objective with a name, owned by your workspace. Agents cannot close their own missions. - Mandate: A revocable, versioned grant that enumerates exactly which actions are allowed, in which currency, for how long, and up to what money ceiling. Authority is a data structure rather than a setting. - Run and ordered steps: Work proceeds as ordered steps against the mandate, each one recorded. A step cannot be skipped and a run cannot silently restart. - Exact-digest approval: High-consequence work requires a single-use approval bound to an exact action digest. If the request that arrives does not match the bytes a human approved, it fails closed rather than proceeding on the strength of a similar-looking one. - Exceptions stop the run: An ambiguous supplier outcome raises an exception, halts the run and waits for an explicit human resolution and resume. A model cannot quietly retry an operation whose result it could not read, which is the failure mode that turns one ambiguous activation into two live campaigns. - Decisions and evidence: What was decided and what proved it are recorded as first-class objects, so the answer to "why did the agent do that?" is a record rather than a reconstruction. - Agents cannot replace or revoke authority. - Agents cannot close missions. - Agents cannot resolve human approvals. - A human credential paired with the reserved MCP marker is rejected rather than silently taking the browser path. - Approval records evidence; it cannot enable spend. Enabling spend is a separately granted action. ### Protocol and transport - Modern MCP, 2026-07-28: Served over both streamable HTTP and stdio. Modern clients negotiate with server/discover, and every later request carries the negotiated version plus client identity and capabilities in its metadata envelope. - Legacy clients still work, and it is tested: Pre-2026 clients may use the legacy initialize handshake over stdio or stateless HTTP. Legacy HTTP returns JSON with no session id and is not silently routed through the modern handler's SSE fallback. The suite pins modern negotiation, the latest legacy negotiation, a 2025-03-26 HTTP round trip, exact tool registration, wire metadata and cancellation in both eras. - Stateless HTTP: The HTTP transport needs no session affinity, so it works behind tunnels and load balancers without sticky routing. - The boundary is fail-closed: Production must set exact allowed hosts and allowed browser origins — `none` for a deployment with no browser clients. Local execution binds to loopback unless a container deployment opts into 0.0.0.0 explicitly. Verifier time, request concurrency and JSON body size are all bounded, and a modern HTTP request that omits its protocol version header is rejected before dispatch, preserving the JSON-RPC id and making no downstream call. ### Why your Claude session is not an agent credential The most consequential design decision in this server is a refusal. A human's ordinary web session cannot be forwarded as an agent's API credential, and the attempt fails with a 401 rather than quietly working. Every request carries a verified actor class that cannot be changed by adding or removing headers. Ordinary human sessions are deliberately classified as human and web; the MCP client's fixed source marker makes that pairing fail rather than silently taking the browser path. An agent credential must be either a configured service token or a verified delegated token carrying a signed agent actor claim and a non-empty verified client id. The API also refuses to let an agent credential resolve a human approval, and it persists actor, source and client attribution on every action. It matters because the alternative — an agent inheriting whatever a logged-in human can do — is how an agent ends up authorising its own spend. - Local development: Run the API with auth disabled and the MCP server needs no environment variables at all. This mode is explicitly not an authentication boundary — it grants its single dev identity every human permission implicitly — and it is labelled that way in the code rather than only in the docs. - Private single-tenant: A service token plus an advertiser id on the API, and the matching token on MCP. Use a distinct random token per tenant-scoped deployment. Suitable for one organisation, never as public multi-user auth. - Public OAuth: Not shipped. A trusted runtime module would supply the integration, and startup already fails if OAuth is mixed with static tokens, but the package deliberately does not invent or embed an authorisation server. Choosing the issuer, client registration, callback allowlist, consent and tenant claim — and a broker or separately reviewed verifier for a non-Clerk issuer — is an open launch gate rather than a configuration step. ### Setting it up against a deployment you control - Claude Code, stdio: claude mcp add adbuymcp -- node /path/to/adbuymcp/apps/mcp/dist/stdio.js — The recommended local path. Point it elsewhere with --env ADBUY_API_URL=... - Claude Code, HTTP: claude mcp add --transport http adbuymcp http://localhost:4200/mcp — Start the server first. Add --header "Authorization: Bearer " if a bearer secret is set. - Claude Desktop, stdio: { "mcpServers": { "adbuymcp": { "command": "node", "args": ["/path/to/apps/mcp/dist/stdio.js"] } } } — Goes in claude_desktop_config.json. Restart Desktop and the tools appear under the hammer icon. - claude.ai, Remote HTTP: Settings → Connectors → Add custom connector → https:///mcp — For a deployment you host and control. Expose the HTTP transport over HTTPS with a bearer secret set, and add the exact hostname to the allowed-hosts list before accepting traffic. There is no hosted AdBuyMCP connector to add, and the unauthenticated server must never be exposed. GET /healthz returns service status and the API URL, with no authentication required. ### Tool annotations, and what they are not - Read-only: 38 - Destructive: 27 - Idempotent: 17 - Open-world: 14 These are discovery hints that help a client decide what to surface and what to confirm. They are not the security boundary. The boundary is the API's fail-closed execution hook, which rejects any consequential call that does not match a registered action and an approval granted for those exact bytes. Q: Can I connect to a hosted AdBuyMCP server today? A: No. What exists is a server you deploy yourself, single-tenant, behind a bearer token or your own tunnel. Public multi-tenant MCP needs an OAuth 2.1 authorisation server issuing per-user, per-organisation, audience-bound tokens, and the package deliberately does not embed one — that is an open gate rather than a setting. The setup instructions below are for a deployment you control. Q: Is MCP the product, or a feature of it? A: A feature, and the site is named after it because it is the feature that changes how the product is used rather than the one that makes it defensible. MCP is an integration surface — Amazon and other advertising platforms already expose one — so an MCP server alone is not a moat. What is unusual here is that the agent transport and the browser are the same execution path, so pointing an agent at a media budget does not mean accepting a weaker set of rules than a person would get. Q: Can Claude spend my money without asking? A: No, and the block is structural rather than a permission you remember to switch off. Consequential calls fail closed unless the incoming request matches an approval a human granted for those exact bytes, spend runs against a mandate with an explicit money ceiling and expiry, and enabling spend on a live line is a separately granted action that approval alone cannot perform. Agents cannot revoke or replace authority, close missions, or resolve human approvals. Q: What happens when a supplier returns something ambiguous? A: The run stops. An ambiguous outcome raises an exception that requires an explicit human resolution before the run resumes, and it is never blindly retried. That is the specific failure this design exists to prevent: an agent that cannot read a supplier's response and tries again has just created a second campaign it does not know about. Q: Which MCP protocol versions do you support? A: The modern 2026-07-28 protocol over both streamable HTTP and stdio, with server/discover negotiation and per-request metadata. The pre-2026 initialize handshake also works over stdio and stateless HTTP, and the compatibility path is pinned by tests rather than assumed — including a 2025-03-26 HTTP round trip and cancellation behaviour in both eras. Q: Can I use this from something other than Claude? A: Yes. It is a standard MCP server over stdio and streamable HTTP, so any MCP-compatible client can drive it. The setup instructions name Claude Code, Claude Desktop and claude.ai because those are the clients most people have, not because the server knows the difference. ## 9. Creative — https://adbuymcp.com/creative Brief to channel-correct creative, with the format rules enforced in code rather than requested in a prompt. Then, if you want them, finished assets on a published ladder: 50p for an image, £1.25 for a voiceover, £15.60 for ten seconds of premium video. The creative studio produces the right shape of creative for each channel from one sentence: responsive search copy inside Google's real character limits, paid social copy bounded to a portable envelope that is valid on both Meta and TikTok, fifteen and thirty second audio scripts with voiceover and SFX direction, DOOH copy rendered to genuine SVG at UK screen resolutions, CTV and cinema storyboards with the trafficking metadata attached, and TV scripts carrying Clearcast substantiation notes. Every one of those limits is a schema constraint, so copy that cannot fit fails at the brief boundary rather than arriving truncated at a provider. ### Formats, with the limits that are enforced rather than advised - Paid search — Responsive search ads. 3 to 15 headlines at 30 characters each and 2 to 4 descriptions at 90 characters, as schema minimums and maximums. Fewer than three headlines is as invalid as a thirty-first character. - Paid social — Bounded social copy. Headline 40 characters, description 30. Conservative portable limits chosen to sit inside TikTok's tighter 100-character text envelope, so one creative is valid on Meta and TikTok rather than needing two. - Podcast & audio — 15 and 30 second scripts. Written with voiceover direction and SFX notes rather than as bare copy, because a script with no read direction is half a brief and the read is most of the ad. - DOOH — Real SVG creatives at UK screen resolutions. 1080×1920, 1920×1080 and 1400×400, rendered rather than mocked up. Headline capped at seven words. Legal copy is held separately from the headline, subline and call to action, because those three are word-clamped and legal text is not. - CTV & streaming — Storyboards with VAST metadata. 16:9 ProRes 422 HQ mezzanine at 1920×1080, 25fps for the UK, at least 20Mbps. Supers inside 90% title-safe, loudness to EBU R128. - Cinema — Storyboards and a DCP delivery spec. 2K flat at 1998×1080, 24fps, 5.1 audio, 60 seconds maximum, routed via DCM Cinemapper with a CAA clearance certificate before print traffic. - Addressable & linear TV — TV scripts with substantiation notes. Every superlative or market-leadership claim is flagged with the substantiation Clearcast will ask for, on the record, before it reaches a clearance queue. ### Clearance, predicted before it costs you a week A pre-check runs every creative against the UK codes that actually decide whether it airs, and returns the findings rather than only the verdict: a blocker fails it, a warning flags it, and each finding names the rule it came from. Its rule checks are deterministic; where a model reads the copy, the finding says which path produced it. It is a prediction of what Clearcast, the CAA and publishers will flag, produced before you commission the asset. - CAP 3.7 / BCAP 3.9: Objective and superlative claims — best, number one, market-leading — need documentary substantiation. The pre-check finds the claim and names it. - BCAP 11: Medicines, medical devices, treatments and health claims. Clearcast requires evidence and may block outright; an unsubstantiated health efficacy claim is routinely rejected. - CAA age-rating: Cinema copy containing strong language restricts pairing to 15 and 18 certificate films, which is a media constraint as much as a creative one. WHAT THE PRE-CHECK CANNOT DO: The pre-check cannot read a substantiation document and does not pretend to. Where you declare that you hold evidence, it records the declaration alongside the claim and says explicitly that it does not assess evidence quality — Clearcast will require the documents themselves before clearance. Both the claim and the declaration stay on the record, which is the useful thing to hand the body that actually decides. ### Finished assets, not just copy Five modalities, generated through the same provider stack, priced from the configured provider cost bases at run time and charged as their own statement line. - Video, via fal.ai: Three tiers: Wan for drafts, Kling v2.5 Turbo Pro for standard, Kling v3 Pro for premium. Five to ten second clips, 16:9, 9:16 or 1:1. - Image, via FLUX via fal.ai: Stills for social, display and DOOH backplates. - Voiceover, via ElevenLabs: UK male, female and neutral voices, at slow, normal or fast pace. - Music, via stable-audio: Beds from 5 to 180 seconds. - Motion, via Programmatic or AI-rendered: 5 to 15 second loops. Programmatic motion is generated SVG and effectively instant; rendered motion is an AI animated-text loop of 5 to 10 seconds. Both output at 1080×1920, 1920×1080 or 1400×400. - Image (FLUX): 50p (at the 50p floor) - Voiceover (ElevenLabs): £1.25 (UK voices, male, female and neutral) - Music track: 50p (at the 50p floor) - Programmatic motion loop: 50p (5–15s SVG, generated instantly) - Rendered motion loop, 10s: £2.00 (AI-rendered animated text) - Video, 10s standard: £6.00 (Kling v2.5 Turbo Pro) - Video, 10s premium: £15.60 (Kling v3 Pro) Priced at 80% gross margin on raw provider cost — retail is cost divided by one minus the margin, so 5× cost, with a 50p floor — debited from the wallet at submit. A distinct hard charge with its own statement line, never blended into the 10% media fee. CAVEATS THAT TRAVEL WITH THOSE PRICES: The 80% is a pricing rule against raw provider cost, not an achieved gross margin: it excludes storage, compute, egress, support and overhead. The cost bases behind the ladder are also explicitly unverified against a live provider invoice, and and we must reconcile them before charging real customers. Prices reprice from configuration, so read the ladder as a current default. Refunds: A definite provider failure refunds the charge automatically. An ambiguous outcome does not: the charge is retained and retry is disabled, so a duplicate render cannot be produced and billed, and it is cleared by a named human with reconciliation rights. Sandbox: Without provider keys, every modality returns a deterministic placeholder rather than an error, so the whole creative loop runs end to end before anyone pays for a render. ### Where the model stops - The model writes. Code does the arithmetic.: Plan generation is grounded in adapter forecasts and every number in a plan is computed deterministically. A language model is not asked to add up a budget, because a model that is wrong about arithmetic is wrong in a way that looks right. - The model never emits a vendor id: It extracts semantics; registry tables emit segment identifiers. A hallucinated segment cannot reach a buy because the model was never holding the pen at that point. - Limits are schemas, not instructions: A prompt asking for thirty characters produces thirty-one often enough to matter. A schema rejecting thirty-one produces a 400 at the edge, every time, on every path. - There is a full deterministic fallback: With no API key the persona extractor, plan generator, pre-check, optimiser and fidelity narrator all run on heuristics instead. The product works without the model; the model makes it better rather than making it possible. AI-produced output is visibly distinguished from deterministic fallback output wherever it appears. The AI layer runs on Claude — claude-opus-5 by default, overridable — for persona extraction, planning, creative pre-check, optimisation and fidelity narration. ### Who owns the output, and who carries the risk Every configured model's licence position is served alongside its price rather than living in a document somewhere. Anything not on that list is reported as unverified rather than permitted. - The default image model is openly licensed: FLUX.1 [schnell] ships as the default under an Apache-2.0 open-commercial licence, so commercial use is permitted by the licence itself rather than by a badge. - Closed-weight models rest on documentation, not a warranty: For the closed-weight video and audio models, commercial-use permission rests on the host's documentation rather than on a contractual warranty. That is a weaker position and it is stated as one. - Output ownership is not granted by the host's contract: The provider's terms do not grant ownership of generated output, and the provider notes that output may not be unique across customers. If exclusivity matters to your brand, this is not the route to it. - There is no IP indemnity — and it runs the other way: The provider offers no intellectual-property indemnity. Under its terms the customer indemnifies the provider rather than the reverse, and output is not warranted original or non-infringing. This is the single most important thing on this page for a legal reviewer, which is why it is not in a footnote. Provider terms read on 12 August 2026; the terms themselves were last updated on 3 March 2026. Q: Will the AI write ads that get rejected? A: It will write ads that are structurally valid, because the format limits are schema constraints rather than instructions, and it will flag the claims most likely to be challenged before you spend anything on production. What it cannot do is guarantee clearance: the pre-check predicts what Clearcast, the CAA and publishers will flag, and those bodies decide. Where you declare that you hold substantiation for a claim, the system records the declaration and states plainly that it has not assessed the evidence. Q: Do I have to use the AI creative? A: No. You can bring finished assets and use the studio only for the channel adaptation and the pre-check, or use it for structure and produce the assets elsewhere. The asset charge only applies to what you actually generate here. Q: Why is ten seconds of premium video £15.60 when competitors include AI creative free? A: Because free creative is not free, it is bundled, and a bundled cost is one you cannot audit. £15.60 is the top of the ladder rather than the price of video: ten seconds of standard video is £6.00. It is a genuinely harder sell — Universal Ads and Global AdPower both include AI creative — and the honest case is that a visible margin on a small render is better for you than an invisible one inside a media percentage. What this page will not do is quote a "versus traditional production" saving to make the number look generous: we deleted our own such comparison as unsourced and added a test to stop it coming back. Q: What happens if the provider fails halfway through? A: It depends on whether the failure is definite. A definite failure refunds the charge automatically to the wallet, as its own ledger entry referencing the failed job. An ambiguous outcome — where the provider may or may not have produced the asset — retains the charge and disables retry, precisely so a second render cannot be produced and billed, and it is resolved by a named human with reconciliation rights. There is no warn-only path for asset charges at all: a real provider cost is incurred at submit, so the wallet check is a hard gate with no override. Q: Who owns the AI-generated assets, and are they indemnified? A: You get the asset and you can use it commercially, and two things are weaker than you might assume. The provider's terms do not grant ownership of generated output and note it may not be unique across customers. And there is no intellectual-property indemnity: under those terms the customer indemnifies the provider rather than the reverse, and output is not warranted original or non-infringing. The default image model, FLUX.1 [schnell], is Apache-2.0 and openly licensed for commercial use; the closed-weight video and audio models rest on the host's documentation rather than a contractual warranty. Each model's position is served next to its price, and an unlisted model reports as unverified rather than permitted. Q: Does AI-generated video need to be disclosed? A: On TikTok, yes, and the platform enforces it rather than leaving it to you. Video produced here is submitted with a self-disclosure provenance declaration, missing provenance blocks the call before any provider request is made, and the provider readback must still carry the same immutable declaration at resume. ## 10. Pricing — https://adbuymcp.com/pricing Ten per cent, and here is everywhere else the money goes. The platform fee is 10% of media spend, disclosed on every line, with no spread and no undisclosed margin anywhere in the path. Above £25,000 of trailing 30-day spend it drops to 5% plus a £749 monthly platform fee. AI asset production is charged separately at a published rate and is never blended into the media fee. Those are all the fees there are, and the rest of this page is about the ones that are not ours: channel floors, data CPM uplifts, and the budget below which none of this is worth doing. ### The two tiers - Standard: 10% of media, platform fee None. Up to and including £25,000 of trailing 30-day media spend. One fee ledger entry accrues per delivery row, carrying the date, the campaign, the channel, the spend and the fee. The fee is visible against the media it was charged on rather than summarised at the bottom of a statement. - Scale: 5% of media, platform fee £749 per month. Strictly above £25,000 of trailing 30-day media spend. The tier is evaluated on trailing 30-day spend, so it switches on when your spend does rather than at a contract renewal. The monthly fee is charged for any month in which the scale tier was active on any day, and it appears as its own statement line. ### Worked examples - £10,000 / month: AdBuyMCP £1,000 (Standard, 10%); modelled agency £1,500. Below the threshold. The standard rate applies to every pound. - £25,000 / month: AdBuyMCP £2,500 (Standard, 10%); modelled agency £3,750. Exactly at the threshold, so the standard rate still applies. The scale tier begins strictly above this figure. - £50,000 / month: AdBuyMCP £3,249 (Scale, 5% + £749); modelled agency £7,500. £2,500 in fee plus the £749 platform fee. - £100,000 / month: AdBuyMCP £5,749 (Scale, 5% + £749); modelled agency £15,000. £5,000 in fee plus the £749 platform fee. WHAT THE COMPARISON COLUMN IS WORTH: Read this column as arithmetic, not as evidence. It models 15% of media with a £750 monthly minimum, which is the assumption our own calculator uses — a calculator that is switched off by default and whose own basis line reads "illustrative only, not a measured benchmark". There is no measured agency rate card behind it. Real agency terms vary widely, plenty are better than this, and the fee is not what an agency is selling: strategy, creative judgement, channel expertise and someone accountable are, and none of them appear in a percentage. Compare it against the terms you actually have. ### AI asset production, charged separately on purpose Finished assets are a distinct hard charge, priced at an 80% margin on raw provider cost — retail is cost divided by one minus the margin, so 5× cost, floored at 50p — debited from the wallet at submit and automatically refunded if the provider fails. It never touches the media fee. Why it is charged separately: Blending asset costs into a media percentage is how an ad-tech margin becomes invisible. Keeping them apart costs a simpler headline and buys a statement where every line can be checked against something. - Image (FLUX): 50p (at the 50p floor) - Voiceover (ElevenLabs): £1.25 (UK voices, male, female and neutral) - Music track: 50p (at the 50p floor) - Programmatic motion loop: 50p (5–15s SVG, generated instantly) - Rendered motion loop, 10s: £2.00 (AI-rendered animated text) - Video, 10s standard: £6.00 (Kling v2.5 Turbo Pro) - Video, 10s premium: £15.60 (Kling v3 Pro) CAVEATS: Two caveats, both from us. The cost bases these prices derive from are unverified — the configuration template says in capitals that they have not been reconciled against a live provider invoice and must be before real customers are charged. And the 80% is a pricing rule computed against raw provider cost, excluding storage, compute, egress, support and overhead, so it is not an achieved gross margin. The ladder reprices from configuration at run time, so treat these as a current default rather than a committed price list. There is deliberately no "versus traditional production" figure here. We used to carry one, deleted it as unsourced, and added a test that fails if it comes back. Repeating it on a website would be reintroducing a number nobody could stand behind. ### Prepaid, and the reason is not our cash flow - £100 minimum top-up: The smallest credit the buyer-facing top-up contract accepts. Top-ups, spend, asset charges, asset refunds and the platform fee are five distinct entry kinds in one ledger, so a statement reads as a sequence of events rather than a total. - A launch reserves the fee as well as the media: The funding check at launch covers the media and the maximum platform fee on it, not just the media, so a campaign cannot go live with enough money for the buy and not enough for the fee. If the wallet is short, the refusal names the exact amount: "wallet is £X short of the launch commitment", rather than a generic decline you have to work backwards from. - Media spend fails closed: A launch that the wallet cannot fund does not launch. The check is enforced by default and callers may only opt into warn-only behaviour for sandbox previews, which is the right way round: a preview that warns is useful, a live buy that warns is a bill. - Asset charges are a hard gate with no override: A real provider cost is incurred the moment an asset job is submitted, so there is no warn-only path for assets at all. If the wallet cannot cover it, it does not submit. - Funds stay held through the pause boundary: On a live line, a reservation is not released until spend has been reconciled to the provider's own invoice or account export, and the final match is rechecked inside the same transaction as the release. Missing, mismatched or stale evidence keeps the money held. - Sandbox credit is not money, and cannot become money: Without a payment key, a sandbox top-up credits the wallet with no payment and no network call at all. Live execution refuses a wallet containing sandbox-minted credit outright, so a sandbox balance can never be spent on real media. - GBP only, and VAT is not modelled: The entire money surface is sterling. VAT is not modelled anywhere in the platform and remains an open business decision rather than a configured rate. ### The awkward part, published rather than discovered #### Below about £10,000 a month, this is the wrong product The published band starts at £2,000, and at that end the maths does not work for either of us. A 10% fee on £2,000 is £200 a month against a buyer who needs the most support, and a seven-channel plan at that budget is operationally and statistically weak — too thin to pace properly and far too thin to power a lift test. If your total monthly media is under £10,000, buy search and social directly, and come back when you have a second channel worth funding. #### The fee is not the whole cost of the media Data segments carry their own CPM uplift — around £2.50 per thousand on Bombora topics and £3.00 on Dun & Bradstreet — and every channel has a plan floor enforced when the plan is built: £100 on search and DOOH, £140 a week on CTV, £250 on audio and social, £500 a week on cinema, £1,000 on addressable TV. Vendor-side minimums are separate and bite later; Sky AdSmart's £3,000 burst is the one most likely to surprise a first plan. Those are the sellers' costs rather than ours, and you see them before you approve rather than in the reconciliation afterwards. #### You cannot fund a live campaign here today No buying rail has reached spend-tested or report-verified on our own evidence ledger, so live spend runs inside a design-partner engagement rather than from a self-serve top-up. No fee has ever been charged to a real customer. The sandbox is complete and free, and the supply page grades every rail. #### There is no published customer outcome yet No case studies, no reconciled outcome study, no named reference. That is where a product recruiting its first design partners actually is. The published advertiser band is £2,000 to £50,000 a month; the honest floor is about £10,000. Q: What does AdBuyMCP cost? A: 10% of media spend, disclosed per line, with no spread. Above £25,000 of trailing 30-day spend the rate drops to 5% and a £749 monthly platform fee applies. AI asset production is charged separately at a published rate — 50p for an image, £1.25 for a voiceover, £6.00 for ten seconds of standard video — and is never blended into the media fee. There is no subscription and no seat fee at the standard rate. What you do fund up front is the wallet, which is prepaid with a £100 minimum top-up, and above the threshold that monthly fee is a fixed charge rather than a usage one. A per-channel plan floor is enforced when the plan is built. Q: Is there a hidden margin on the media? A: The commitment is that there is none: the fee engine is built with no spread, one ledger entry accrues per delivery row carrying that row's date, campaign, channel, spend and fee, and the terms commit to media passed through at cost with no undisclosed margin, rebate or arbitrage. Being precise about the status of that: it is a contractual commitment and a design property, not yet a demonstrated practice, because no rail has ever spent and therefore no invoice has ever been reconciled against it. The one place a margin does exist is AI asset production, at an 80% margin on raw provider cost — five times cost — published and charged as its own line. Q: Why is asset production charged separately rather than included? A: Because bundling it is how a margin becomes invisible. It is a genuinely harder sell — Universal Ads and Global AdPower both include AI creative — and the honest case is that a separately charged line you can audit is worth more than a free one you cannot. What this page will not do is quote a saving against traditional production to make £15.60 look generous: the platform carried such a figure, deleted it as unsourced, and added a test to stop it returning. There is no sourced production-cost comparison here. Q: What is the smallest sensible budget? A: Around £10,000 a month across everything, and the published floor of £2,000 is lower than that for reasons of packaging rather than sense. At £2,000 the fee is £200, the support requirement is at its highest, and a multi-channel plan cannot carry enough weight on any single line to pace or to prove anything. Two channels funded properly beats a token line on all seven. Q: Do I pay for the sandbox? A: No. The whole loop — brief, persona, seven compiled plans, creative, approval, launch, delivery, measurement, journeys, optimisation — runs with no credentials and no card, and the data it returns is realistic and repeatable rather than a scripted demo. Charges begin at real media and at real asset generation, both of which incur a real cost to somebody. Q: What happens to my money if a supplier fails? A: It stays held. A wallet reservation is only released once spend has been reconciled to the provider's own invoice or account export, and that final match is rechecked in the same transaction as the release. An asset charge is refunded automatically on a definite provider failure — but an ambiguous outcome is different: the charge is retained and retry is disabled, to stop a duplicate render being produced and billed, and it is cleared by a named human with reconciliation rights rather than automatically. An ambiguous supplier outcome is never retried blindly either; it stops the run. ## 11. Who this is for — https://adbuymcp.com/solutions Four packages, ranked by how ready each one actually is, with the blocking gaps named on every page. None of them is a proven segment. One cohort is being recruited now, two are ranked behind it, and the fourth is deliberately deferred. Each page carries its own standing. A blank seven-channel canvas is not a product, it is a form. What you want is a shape that already knows your budget floors, your likely channels, your creative checklist and whether your campaign can support a causal claim. These four are the shapes worth building. Publishing where each one is unfinished costs some of the sales it would otherwise win, and buys the ones where you would have found out in week two anyway. ### Who this is deliberately not for - Advertisers under £10,000 total monthly media: A seven-channel proposition cannot be funded or measured at that level, and a 10% fee on it cannot fund the support you will need. 72% of UK SMEs spend under £5,000 a year on digital advertising: they are a real market and this is the wrong product for it. - Enterprises with established trading desks: Global procurement, mature DSP contracts and an in-house desk mean the neutral layer is duplicating something you already own and run better. - Agencies with sophisticated AV or programmatic operations: You already have the capability this replaces. The agency edition, which would need parent and client tenancy, white label, net pricing and non-circumvention, is deliberately deferred rather than half-built. - App-performance buyers: This has no mobile measurement partner, no SKAdNetwork support and no app-network depth. Nothing here would serve you. - High-risk regulated categories: Until operational compliance is proven with a design partner, taking on a category where a clearance failure is a regulatory event would be reckless with your licence rather than ours. - Anyone who needs a causal claim their spend cannot power: If the geography, spend or conversion volume cannot support a valid test, we will tell you so and decline to run it. If a lift number is the deliverable regardless, buy it somewhere that will supply one. Q: Why do your solution pages list what does not work? A: Because the gaps are all things a design partner discovers in the first fortnight, and a supplier who named them in advance is in a completely different position from one who was found out. None of these four packages is a proven vertical: three are hypotheses being validated with design partners and the fourth is deliberately deferred, and pretending otherwise would waste the first meeting on both sides. Q: Do you work with agencies? A: Not yet, deliberately. An agency edition needs parent and client tenancy, permissions, client approval links, net pricing and markup controls, white-label exports, consolidated billing and non-circumvention terms, and none of that exists today. Building it half-way while also running a direct motion would undermine both, so the direct route was chosen and the agency edition was deferred rather than fudged. Q: What if none of these four describe me? A: Then say so on a call, and one of two things is true. Either the shape is close enough that the seven channels and the fidelity model still serve you, in which case a working session will establish it quickly. Or it is not, in which case the useful outcome is finding that out in forty-five minutes rather than after a pilot. ### B2B account surround — https://adbuymcp.com/solutions/b2b-account-surround Reach the whole buying committee beyond LinkedIn. This is the design-partner cohort AdBuyMCP is built for first. A UK B2B business with a known account list and a working LinkedIn and search programme has a specific problem: the buying committee is six to ten people, LinkedIn reaches the two who are on it during the working day, and there is no obvious second channel that reaches a CTO at home. CTV is that channel — consent-chained B2B segments from Bombora and Dun & Bradstreet, delivered against streaming households, and graded person-level at 85 out of 100 where at least one of those segments matches strongly, dropping to household level and to 65 or 60 where they do not. The audience-translation story solves a problem this buyer can already see: a job title cannot be targeted exactly on a television or a billboard, and they would rather be told the score than sold the fiction. Who this is: - 50 to 500 employees, roughly £10m to £250m revenue - 4 to 20 marketers, and no TV, audio or out-of-home trading team - £15,000 to £50,000 available for a two or three channel pilot - High contract value, a CRM, a known account list - Active LinkedIn, search and events programmes already running - Buyers concentrated in UK business centres, sectors or a finite account cluster The job to be done: Reach the whole buying committee beyond LinkedIn, without losing pipeline accountability. Why this product fits: High customer value justifies a measured test, and the concentration of UK B2B buyers in a handful of city centres is exactly the shape that makes both addressable targeting and a geo-lift design workable. The fidelity story is not a caveat here, it is the pitch: this buyer has been told by four vendors that they can target IT decision makers on a billboard, knows it is not quite true, and has no way to compare how untrue each one is. Channels that usually carry it: CTV & streaming, Podcast & audio, Addressable & linear TV, Paid search. WHAT IS MISSING TODAY, which is the half of this page that matters: - No HubSpot or Salesforce connector yet, and no account-list ingestion. Outcome data comes back through analytics and imported outcomes rather than through your CRM. - No account-level outcome reporting. You will see campaign outcomes, not account penetration. - B2B pipeline windows are longer than a four-week flight, so the first campaign will usually close on response evidence rather than on closed-won. - No published case evidence in this segment. Nobody has done this here yet. Where it sits in the priority order: First. This is the cohort being recruited now, explicitly as design partners rather than as customers of a finished performance product. There is no support SLA, no escalation path and no status page yet; what there is, is direct access to the people building it. Q: How is this different from LinkedIn ads? A: LinkedIn reaches the members of the committee who are on LinkedIn, at work, in a professional frame of mind, and it does that better than anything else. What it cannot do is reach the same people at home on a streaming service, in the car on a podcast, or walking through a rail station. This is the surround, not the replacement, and the reason it needs a common audience definition and one exposure ledger is that otherwise you cannot tell whether the surround reached anyone LinkedIn had not already. Q: Can you target my account list? A: Not yet, and that is a real gap rather than a roadmap flourish. Account-list ingestion and CRM connectors are the top blocking item for this segment. Today you define the audience semantically — sector, seniority, firmographics, geography — and it compiles into segments built from the same underlying data. If matching a named account list is the whole requirement, wait for the connector. ### Multi-location growth — https://adbuymcp.com/solutions/multi-location-growth Plan and prove local media across every catchment. Postcode, venue and geographic targeting map onto what we do more naturally than any other use case, because three of the seven channels are bought geographically in the first place: addressable TV by postcode district, DOOH by polygon and venue type, cinema by catchment. The measurement fit is even better. Geo holdouts are the one causal design that works without a login, a pixel or a match rate, and a brand with twenty catchments has enough independent markets to power a test where a national brand with one campaign does not. Who this is: - 20 or more locations, or explicit regional catchments - 50 to 500 employees, a 3 to 15 person marketing team - £15,000 to £75,000 monthly paid media - Usable booking, lead, footfall, point-of-sale or CRM signals The job to be done: Coordinate regional growth across locations, and prove which of it worked. Why this product fits: The channels are geographic and so is the evidence. A matched-market geo-lift test needs comparable markets that behaved alike before the media started, and a twenty-catchment estate supplies them. This is the segment where the phrase "powered causal" is most often achievable rather than aspirational. Channels that usually carry it: Digital out-of-home, Addressable & linear TV, Cinema, Paid search. WHAT IS MISSING TODAY, which is the half of this page that matters: - No store or catchment ingestion. Catchments are defined by postcode district rather than imported from your estate. - No local creative templating, so a per-location creative variant is manual. - No franchise permission model and no per-location budgets. - No call, booking or point-of-sale connectors, and no footfall integration. Outcomes are imported rather than streamed. Where it sits in the priority order: Second, and the strongest vertical package once the ingestion work lands. The measurement fit is better here than anywhere else on this list. Q: Can you prove a billboard drove footfall? A: With enough catchments and enough spend, a geo-lift test can measure whether exposed markets outperformed matched held-out markets on whatever outcome you can supply — bookings, leads, point-of-sale. Without enough of both, the power analysis returns an underpowered verdict and the number of days it would need, and you get response evidence instead. Which of those two happens is decided by your geography and spend, not by us, and it is worth knowing before you fund the campaign. ### New brand launch — https://adbuymcp.com/solutions/new-brand-launch Buy attention before anyone is searching for you. Search cannot harvest demand that does not exist yet, which is the specific bind a new brand is in: the channels that convert cheapest are the ones nobody is using to look for you. A launch buys attention instead — cinema, CTV and DOOH, where a full screen with the lights down or a household streaming in the evening delivers an attention profile the feed cannot. AdBuyMCP's contribution is that the same audience definition drives all three and lands in one exposure ledger, so "how many people did we actually reach, once?" has an answer rather than three overlapping ones. Who this is: - A brand or product with little or no existing search demand - A launch window with a real date attached - Budget concentrated into weeks rather than spread across a year - Willingness to be measured on reach and response rather than on last-click The job to be done: Create demand where none exists, and know how much reach was actually bought. Why this product fits: Cross-channel frequency deduplication matters more at launch than at any other time, because a launch buys the same people repeatedly across channels by design and the naive sum of three channels' reach is the number most likely to end up in a board pack. Cinema is also at its most defensible here: it is a small, volatile market and a distinctive attention product, which is a launch argument rather than a recurring-spend argument. Channels that usually carry it: Cinema, CTV & streaming, Digital out-of-home, Paid social. WHAT IS MISSING TODAY, which is the half of this page that matters: - Causal evidence is usually out of reach at launch: no baseline, a short window, and low conversion volume. Expect reach and response evidence, and say so to whoever is funding it. - Brand lift, which is the natural measure for a launch, is the one measurement verb not yet built here. - Cinema and addressable TV are booking workflows with lead times, so a launch date has to be worked backwards from clearance rather than from the flight. Where it sits in the priority order: Third. Genuinely well-suited to the channel mix, and held back by the fact that the measurement a launch most wants — brand lift — is the placeholder verb. Q: You say brand lift is not built. Why pitch a launch package? A: Because reach, frequency and response are still measurable, deduplicated across channels, and better than what a launch usually gets. What you should not do is fund a launch here expecting a recall study, and a page that let you assume otherwise would be setting up a disappointment for the one moment when you can least afford it. ### Beyond Meta and Google — https://adbuymcp.com/solutions/beyond-meta-and-google Graduate beyond Meta and Google without giving up performance accountability. This is the most crowded position on this page and the page says so. MNTN, Vibe, StackAdapt, retail media and the platforms themselves all sell incremental reach to saturated DTC brands, several of them with better commerce loops and a published incrementality case. What AdBuyMCP offers that they do not is a single audience definition spanning premium channels and the feed together, one exposure ledger to deduplicate them, and a causal engine that refuses to produce a lift number your spend cannot support. If you want a polished performance-CTV product today, buy one of theirs. Who this is: - Post product-market fit, £20,000 to £100,000 monthly paid media - Shopify, GA4 and conversion APIs in place - Reusable creative and a working creative process - Meta and Google saturated: rising costs, flat incremental return The job to be done: Find reach the feed is not delivering, without giving up performance accountability. Why this product fits: The honest fit is the measurement discipline rather than the supply. A saturated DTC brand's real question is whether the next channel is incremental or is buying people who would have converted anyway, and that is a causal question that most platforms answer with attribution. This one answers it with a powered geo test or declines to answer it. Channels that usually carry it: CTV & streaming, Podcast & audio, Digital out-of-home, Paid social. WHAT IS MISSING TODAY, which is the half of this page that matters: - No production Shopify integration and no server-side conversions yet. - No new-versus-existing customer reporting, which is the reporting cut this segment actually runs on. - No publishable incrementality case study. - Competitors here are strong, well-funded and shipping. This is a considered second choice today, not a first one. Where it sits in the priority order: Fourth, and deliberately deferred. Attractive, crowded, and needing product work this cohort would notice the absence of on day one. Q: Why would I not just use MNTN or Vibe? A: For performance CTV on its own today, you probably should — they have a clearer performance story, commerce integrations and polished creative. The case for coming here instead is if you want CTV planned against the same audience as your DOOH, audio and social, deduplicated in one ledger, with a causal engine that will tell you when the test cannot support the claim. If you only want CTV, that machinery is overhead. ## 12. How it compares — https://adbuymcp.com/compare 15 honest comparisons, each one naming the buyer who should choose the alternative. Several of them will lose us a sale, which is the point: a comparison page with no losing case is one you would discount entirely. Q: Who is AdBuyMCP's closest competitor? A: StackAdapt, on capability, and it is ahead on almost every axis that matters today: live supply, a first-party data hub, DCO, attribution, agency tooling and support. The defensible gap is narrower than a category pitch would suggest — UK cinema and addressable television workflows, seller-neutral allocation, published fidelity and policy provenance, and measurement that keeps delivered, responded, caused and remembered apart, with the fourth published as a placeholder rather than a feature. Q: Isn't an MCP server the differentiator? A: No, and claiming it would be the first thing a technical buyer discounted. MCP is an integration surface, and Amazon among others already exposes advertising capabilities through it. What is unusual here is not that an MCP server exists but that the agent transport and the browser exercise the same routes, tenant boundary, wallet gates and approval records, so pointing an agent at a budget does not mean accepting weaker rules. Q: What can you do that nobody else can? A: Take one audience definition and show what it becomes in the native targeting vocabulary of seven UK channels, with a fidelity score and a lawful-basis manifest on each, including the channels that transact as paperwork rather than as APIs — and then keep delivered, responded, caused and remembered apart in the report, with modelled evidence never presented as observed. Each individual piece exists somewhere. The combination, UK-first, with the approximation published rather than implied, is the thing. Q: Why publish the claims you have retired? A: Because anyone who has read four competitor sites has heard every one of them, and the fastest way to be believed about anything is to be visibly unwilling to say the things that are not true. Every phrase on that list was considered for this site and rejected for the reason printed beside it. ### AdBuyMCP vs StackAdapt — https://adbuymcp.com/compare/stackadapt Verdict: They are the stronger DSP by a distance. The gap they leave is UK-specific paperwork and visible approximation. StackAdapt plans, creates, activates, optimises and measures across CTV, audio, DOOH, video, display and native, with its own AI assistant. It has live supply at scale, a first-party data hub, dynamic creative optimisation, attribution, agency tooling, an API and AI video. AdBuyMCP buys through StackAdapt as one of its rails, which is a fair summary of the power balance. What StackAdapt does not do is UK cinema, UK addressable and linear television workflows, or explicit per-channel fidelity scoring and lawful-basis provenance — and a DSP has little reason to prioritise any of those. What they promise, in their words: Plan, create, activate, optimise and measure CTV, audio, DOOH, video, display, native and more, with an AI assistant. WHERE THEY ARE GENUINELY AHEAD: - Live supply at scale, contracted and proven, today - A first-party data hub and dynamic creative optimisation - Attribution tooling and agency workflow features - Support, service and a track record with real clients - AI video generation inside the platform What is actually left for AdBuyMCP: - UK cinema, which is a booking workflow rather than a DSP line - UK addressable and linear television, likewise - An explicit fidelity score per channel, published rather than implied - A lawful-basis manifest and targeting decision history attached to every compiled plan - Neutral allocation across sellers that compete with each other, including StackAdapt itself Choose StackAdapt when: You want a proven programmatic desk today, with live supply, support and a service team, and your channel set is the one a DSP already covers. That describes most buyers, and StackAdapt is the better answer for them right now. Choose AdBuyMCP when: Your plan has to include UK cinema or addressable TV alongside programmatic, or you need the approximation in your targeting written down and auditable rather than assumed. Sources, all from their own published material: - StackAdapt platform: https://www.stackadapt.com/platform ### AdBuyMCP vs Universal Ads — https://adbuymcp.com/compare/universal-ads Verdict: For UK television alone, go direct. AdBuyMCP's argument only starts when a second channel is involved. Universal Ads put ITV, Sky and Channel 4 into one self-serve campaign with no platform tax and included AI creative, and that removed most of the UK television access friction that a neutral layer would have been valuable for. It is a direct seller with direct premium supply, speed, a conversion pixel and measurement partners. We are not going to out-TV them and do not try to: the Universal Ads integration is a rail here, currently a typed stub blocked on partner approval and per-request signing. The remaining argument is seller-neutral allocation and the six other channels. What they promise, in their words: ITV, Sky and Channel 4 in one self-serve campaign, with no platform tax and AI creative included. WHERE THEY ARE GENUINELY AHEAD: - Direct premium TV supply, with no intermediary at all - No platform tax on the media - Speed from signup to a live TV campaign - A conversion pixel and named measurement partners - Included AI creative What is actually left for AdBuyMCP: - Seller-neutral allocation, including deciding how much should go to TV at all - The six non-TV channels, planned against the same audience - One exposure ledger, so TV deduplicates against CTV and DOOH rather than being counted separately - Cinema, which Universal Ads does not sell Choose Universal Ads when: Television is the campaign. If you want ITV, Sky and Channel 4 and nothing else, going direct costs less and starts sooner, and there is no honest argument for adding a layer. Choose AdBuyMCP when: TV is one line in a plan that also has to include audio, out-of-home, cinema, search or social, and you want the split between them argued rather than assumed. Sources, all from their own published material: - Universal Ads UK launch: https://www.itv.com/presscentre/presscentre/presscentre/media-releases/comcasts-universal-ads-launches-uk-channel-4-itv-sky-simplify-access-premium-tv - Universal Ads proposition: https://www.universalads.com/why-ua ### AdBuyMCP vs Global AdPower — https://adbuymcp.com/compare/global-adpower Verdict: They own the inventory. If your audio and outdoor plan is a Global plan, buy it from Global. Global's AdPower put self-serve radio, digital audio and outdoor behind a signup, with free AI creative and next-day launch, on inventory Global owns directly: Heart, Capital, LBC, TfL, bus and street hubs. That is a genuinely strong proposition and it removed a lot of the friction a neutral layer would have been useful for. What it cannot do, structurally, is plan across sellers — a Global product will not recommend a competitor's screen — or carry a common audience and measurement model across television, cinema and the feed. What they promise, in their words: Self-serve radio, digital audio and outdoor, with free AI creative and next-day launch. WHERE THEY ARE GENUINELY AHEAD: - Direct ownership of Heart, Capital, LBC, TfL, bus and street-hub inventory - Next-day launch - Free AI creative - No intermediary fee on their own supply What is actually left for AdBuyMCP: - Cross-seller planning, including sellers competing with Global - Television and cinema, which AdPower does not sell - A common audience definition and one measurement model across all of it Choose Global AdPower when: Your audio and outdoor buy is Global inventory, and you want it live tomorrow. Going direct is faster and cheaper, and we would be a layer between you and something you could reach yourself. Choose AdBuyMCP when: You need audio and outdoor bought alongside television, cinema, search and social, against one audience, deduplicated in one ledger. Sources, all from their own published material: - Global AdPower launch: https://global.com/global-launches-adpower-self-service-advertising/ ### AdBuyMCP vs A media agency — https://adbuymcp.com/compare/media-agency Verdict: The fee comparison flatters us and the capability comparison does not. Both are true. At £100,000 a month, our fee is £5,749 against £15,000 under a 15%-plus-minimum model, and the arithmetic is real. The model is not: there is no measured agency rate card behind that 15%, our own calculator carries the basis line "illustrative only, not a measured benchmark", and it is switched off by default. It is also the least interesting part of the comparison. An agency brings strategic judgement, creative quality, channel expertise built over years, someone accountable who answers the phone, reconciliation, and the procurement readiness that gets a supplier through a large company's onboarding. This has software, a published fee and an audit trail — and no account team, no support SLA and no status page, because those are unbuilt rather than unmentioned. For a marketing team with no media capability of its own, an agency is often still the right answer. What they promise, in their words: Managed expertise, plus planning, operations, billing and reporting. WHERE THEY ARE GENUINELY AHEAD: - Human strategic judgement, and a person accountable for the outcome - Creative quality and channel expertise built over years - Client service, and someone to ring when it goes wrong - Reconciliation and procurement readiness - A track record you can take up references on What is actually left for AdBuyMCP: - A published fee, and a contractual commitment to no undisclosed rebate anywhere in the path - The fee visible on the line of media it was charged on, rather than in a quarterly summary - An audit trail of every targeting decision, with its lawful basis - A causal engine that refuses underpowered tests rather than reporting one - A lighter operating layer for teams that have judgement but no trading desk Choose an agency when: You have no in-house media capability and need judgement more than you need tooling, or your procurement process requires a supplier with references and an insurance profile we do not yet have. Choose AdBuyMCP when: You already have the judgement, you want the trading and reconciliation layer without the retainer, and you want every fee visible against the media it was charged on. Sources, all from their own published material: - IPA / Tracksuit client-agency research: https://pre.ipa.co.uk/news/bridging-the-gap/ - IPA Agency Census 2025: https://ipa.co.uk/news/agency-census-2025/ - UK PPC agency fee survey: https://finsburymedia.com/how-much-do-ppc-agencies-charge/ - UK agency retainer research: https://whito.co.uk/research/uk-agency-retainers/ ### AdBuyMCP vs Google and Meta — https://adbuymcp.com/compare/google-and-meta Verdict: Not a replacement. An incremental layer around the budget you are already spending well. Search was £17.9bn and social £11.5bn of a £46.7bn UK market in 2025, which is 63% between them, and they earned it: familiar, liquid, closed-loop, optimised daily, with a low operational burden. Nothing here is going to beat Google at Google. AdBuyMCP buys both — Google Ads and Microsoft for search, Meta and TikTok for social — precisely so that a plan can be honest about incremental reach. The one thing this offers against the duopoly is reach beyond the feed with a measurement model that will tell you when the incrementality claim is not supportable. What they promise, in their words: Familiar, liquid, closed-loop acquisition where most of the budget already is. WHERE THEY ARE GENUINELY AHEAD: - Most of your budget already lives there, and it works - Existing data, existing conversion signals, existing pixel - Daily optimisation nobody outside the platform can match - A low operational burden What is actually left for AdBuyMCP: - Incremental reach beyond the feed, if it can be proved - One audience definition spanning the feed and the premium channels around it - Cross-channel frequency deduplication, so a reach number is not three overlapping ones - A causal test that refuses to run when it cannot answer the question Choose Google and Meta when: Your acquisition is still scaling profitably inside search and social. Adding channels before the cheap ones are saturated is a way to spend more for less, and we will not tell you otherwise. Choose AdBuyMCP when: Costs are rising and incremental return is flat, and the question has changed from "how do we spend more here?" to "is there anyone left here we have not already reached?" Sources, all from their own published material: - AA/WARC 2025 UK expenditure: https://adassoc.org.uk/our-work/aa-warc-updates-advertising-expenditure-report-to-capture-evolving-uk-media-landscape-and-records-46-7bn-media-investment-in-2025/ - IAB UK 2025 digital expenditure: https://www.iabuk.com/news-article/digital-adspend-2025-uks-digital-ad-market-reaches-ps405bn ### AdBuyMCP vs Doing it yourself — https://adbuymcp.com/compare/doing-it-yourself Verdict: For one or two channels, do it yourself. The case for a layer starts at three and gets stronger with each one. Nothing here is inaccessible. Universal Ads sells you TV, Global sells you audio and outdoor, DCM sells you cinema, Google and Meta sell you the feed, and a DSP seat sells you programmatic. If your plan is one or two channels, buying direct is cheaper and faster, and we would be a fee between you and a thing you could reach yourself. The cost that grows is not the buying. It is maintaining seven definitions of the same audience in seven systems, none of which can tell you whether they overlap, and reconciling seven reports into one number at the end of the month. What they promise, in their words: No platform fee, and a direct relationship with every seller. WHERE THEY ARE GENUINELY AHEAD: - No layer fee at all - Direct relationships and direct support from each seller - Full control, and no dependency on a third party's roadmap - For one or two channels, genuinely less work What is actually left for AdBuyMCP: - One audience definition instead of seven that drift apart - Cross-channel frequency deduplication, which no single seller can do - One report rather than seven exports and a spreadsheet - Cinema and addressable paperwork automated rather than emailed - A causal test spanning all of it rather than per-channel attribution that double-counts Choose doing it yourself when: You are buying one or two channels, or you have a trading desk and the people to run it. The overhead of a layer is only worth it when there is enough to coordinate. Choose AdBuyMCP when: You are running three or more channels, spending time reconciling them by hand, and cannot currently answer how many people you reached once across all of them. Sources, all from their own published material: ### AdBuyMCP vs MNTN — https://adbuymcp.com/compare/mntn Verdict: They have the proven performance-CTV product. We publish what we charge. Decide which of those you need more. MNTN built the performance-television category: self-serve connected TV with outcome measurement, Verified Visits attribution and a creative marketplace, listed on the New York Stock Exchange in May 2025 on 2024 revenue of $226m. It is a real product with real scale, and we are not going to pretend otherwise. The comparison turns on one thing. MNTN's own pricing documentation states that it keeps a percentage of media spend and does not disclose its margin — a position it has been criticised for consistently in the trade press. AdBuyMCP charges a published 10%, falling to 5% plus £749 a month above £25,000 of trailing thirty-day spend, with one fee ledger entry accruing per delivery row so the fee sits beside the media it was charged on. If you are buying connected TV alone and you do not mind an undisclosed take, MNTN is the more finished product today. What they promise, in their words: Performance TV: self-serve connected TV with ROAS measurement, Verified Visits attribution and a creative marketplace. WHERE THEY ARE GENUINELY AHEAD: - A proven performance-CTV product with real customers and public financials - Verified Visits attribution, built and validated over years - A creative marketplace attached to the buying platform - Scale and momentum: $226m 2024 revenue, and growing at 25 to 48% year on year through 2025 - A single-channel product is simpler to run than a seven-channel one What is actually left for AdBuyMCP: - A published take rate rather than an undisclosed margin inside the price - Six channels MNTN does not sell, including UK cinema and addressable television - UK-first supply and UK clearance workflows - A causal engine that refuses underpowered tests rather than reporting one Choose MNTN when: Connected TV is the campaign, you are buying US inventory, and you want a product that has been proven with real advertisers. Their effective programme minimums — around $10,000 for retargeting and $25,000 for prospecting on third-party comparisons — put them out of reach below that, but above it they are the established choice. Choose AdBuyMCP when: You need to know what you are being charged, or connected TV is one line in a UK plan that also has to carry audio, out-of-home, cinema, search or social against the same audience. Sources, all from their own published material: - MNTN pricing model: https://help.mountain.com/en/articles/8054978-understand-mntn-s-pricing-model - MNTN Q2 2025 results: https://s205.q4cdn.com/886349678/files/doc_news/MNTN-Reports-Strong-Second-Quarter-2025-Results-2025.pdf ### AdBuyMCP vs tvScientific — https://adbuymcp.com/compare/tvscientific Verdict: Their pricing model is better than ours for a qualifying performance buyer. That is not a concession, it is the honest answer. tvScientific sells performance connected TV on cost-per-outcome: you pay when the defined outcome happens — an install, a purchase, a visit — with the platform carrying the delivery risk and its margin embedded in the outcome price. It also offers Guaranteed Outcomes and runs self-serve with no minimum spend. It raised a $25.5m Series B in February 2025 with Roku among its investors. Stated plainly: for an advertiser who qualifies and whose objective is a measurable direct outcome, paying per outcome is a better commercial structure than paying a percentage of media, and no amount of transparency on our side changes that. Outcome pricing requires measurement maturity and risk capital, and we do not offer it. What they promise, in their words: Performance connected TV bought on cost-per-outcome, with guaranteed outcomes and no minimum spend. WHERE THEY ARE GENUINELY AHEAD: - Cost-per-outcome pricing, the most aggressive pricing innovation in this category - No minimum spend on self-serve - Patented outcomes measurement built for the model - Named customers across retail, beauty, gaming and consumer What is actually left for AdBuyMCP: - Six channels they do not sell, and UK-specific booking workflows - Brand-side planning, which their tooling is thin on by their own positioning - A seller-neutral layer across competing supply rather than one rail - UK-first supply: they are US-centric Choose tvScientific when: You are a performance advertiser with a clean, high-volume conversion event, you qualify for their outcome pricing, and connected TV is the channel you want. On those facts their model transfers risk away from you in a way a percentage of media does not, and you should take it. Choose AdBuyMCP when: Your objective is not a single trackable outcome, or connected TV is one line in a UK plan spanning channels they do not sell. Sources, all from their own published material: - tvScientific launches CPO: https://www.tvscientific.com/press/press-releases/tvscientific-launches-cpo - Guaranteed Outcomes: https://www.tvscientific.com/press/press-releases/guaranteed-outcomes ### AdBuyMCP vs The Trade Desk — https://adbuymcp.com/compare/the-trade-desk Verdict: The category's most capable buying platform, priced accordingly. Not a fair fight on supply, and not the same job. The Trade Desk is the largest independent demand-side platform and is ahead of AdBuyMCP on essentially every axis a trading desk cares about: inventory, identity resolution, verification, optimisation and scale. The comparison exists for one reason, which is price and simplicity. Its platform fee was about 20.3% of spend in 2024 on published analysis, with one to two point discounts negotiated at $500,000 to $750,000 thresholds — a rate that is competitive at enterprise volume and punitive below it. We are not an alternative to it for a trading desk. It is an alternative for a marketing team that does not have one and does not want to build one. What they promise, in their words: Enterprise omnichannel execution with deep supply, identity, data and optimisation. WHERE THEY ARE GENUINELY AHEAD: - Inventory access and supply relationships at a scale nothing here approaches - Identity resolution and its own industry identifier standard - Verification, brand safety and fraud controls, none of which exist here - Optimisation built and tuned over more than a decade - A trading interface built for people who trade media full time What is actually left for AdBuyMCP: - A published fee at any volume, rather than a rate negotiated against spend thresholds - UK cinema and addressable television, which are paperwork rather than programmatic - A neutral business layer above competing sellers rather than one platform's seat - Simplicity for a team without a trader Choose The Trade Desk when: You have a trading desk, or an agency running one for you, and enough volume to negotiate. At that scale their capability is not comparable to anything here and the fee buys something real. Choose AdBuyMCP when: You have no trader, your channel set includes UK cinema or addressable television, and you want the approximation in your targeting written down rather than assumed. Sources, all from their own published material: - Analysis of The Trade Desk's take rate: https://www.adexchanger.com/online-advertising/as-ctv-blooms-its-knives-out-for-the-trade-desks-take-rate/ ### AdBuyMCP vs Display & Video 360 — https://adbuymcp.com/compare/dv360 Verdict: Unmatched scale inside Google's ecosystem. The neutrality question is the whole point. Display & Video 360 is Google's enterprise buying platform, with a platform fee usually quoted at around 10 to 15% of spend. It is deeply integrated with the rest of the Google stack, which is its strength and the reason to think carefully. A buying platform owned by the largest seller of advertising in the world has a structural interest that a neutral layer does not, and that is not an accusation — it is the reason seller-neutral allocation is a category at all. AdBuyMCP buys Google Ads as one rail among twenty-five and has no stake in how much of your budget goes there. What they promise, in their words: Enterprise omnichannel buying, deeply integrated with Google's data, inventory and measurement. WHERE THEY ARE GENUINELY AHEAD: - Scale and inventory access, particularly on YouTube and Google's own supply - Integration with Google Analytics, Ads and the wider stack - Mature audience, frequency and measurement tooling - Enterprise support and agency familiarity What is actually left for AdBuyMCP: - Seller neutrality: we have no stake in where the budget lands - UK cinema and addressable television workflows - A published fidelity score on every compiled audience - Accessibility below enterprise volume Choose DV360 when: Your media is predominantly digital, YouTube matters to you, and you already run inside Google's stack. The integration is genuinely valuable and duplicating it through a layer would be daft. Choose AdBuyMCP when: You want allocation decided by someone with no stake in the answer, or your plan includes channels Google does not sell. Sources, all from their own published material: - DSP fee benchmarks: https://improvado.io/blog/dsp-vs-ssp-programmatic-guide ### AdBuyMCP vs Amazon DSP — https://adbuymcp.com/compare/amazon-dsp Verdict: We are built to buy through them. If Amazon audiences and CTV are the whole plan, go direct and skip the layer. This comparison is awkward and the honest thing is to say why. Amazon DSP is one of the twenty-five connectors AdBuyMCP registers — a sub-seat covering connected TV and the audio pipe — and it is also a direct alternative to using AdBuyMCP at all. Both are true. Amazon's advantage is commerce data no third party can see and inventory across its own properties. Its access is not open: a sub-seat comes through an authorised reseller rather than a signup, which is a friction a small advertiser feels. If your plan is Amazon audiences against Amazon and streaming inventory, buying it directly costs less and starts sooner, and there is no honest argument for a layer between you and it. What they promise, in their words: Amazon's audiences and inventory, including Prime Video and the wider commerce ecosystem. WHERE THEY ARE GENUINELY AHEAD: - Commerce and purchase data no third party can replicate - Owned inventory across Prime Video and Amazon's properties - Closed-loop measurement inside their own ecosystem - Scale, and a direct relationship without an intermediary What is actually left for AdBuyMCP: - The five channels Amazon does not sell, planned against the same audience - One exposure record, so Amazon deduplicates against everything else rather than being counted separately - Seller-neutral allocation, including how much should go to Amazon at all - UK cinema and addressable television Choose Amazon DSP when: Amazon audiences and inventory are the plan, and you can get a sub-seat. Adding a layer to reach one seller you can reach directly is a fee for nothing. Choose AdBuyMCP when: Amazon is one leg of a UK plan and you want the split between it and five other channels argued rather than assumed — in which case Amazon DSP is one of the rails AdBuyMCP is built to buy through. Sources, all from their own published material: - Amazon DSP: https://advertising.amazon.com/solutions/products/amazon-dsp ### AdBuyMCP vs Blis — https://adbuymcp.com/compare/blis Verdict: Genuinely strong in the UK and on privacy, with a fee structure that beats ours. The gap is addressable TV, cinema and published fidelity. Blis is a privacy-first plan, buy and measure platform with real UK geographic scale, an audience explorer, holdout-based measurement and — the part worth taking seriously — no media markup and no platform fee. On the two things we argue hardest about, transparency and privacy posture, Blis is not the soft target the category's opaque players are. What remains is narrower than a category pitch would suggest: UK addressable television and cinema workflows, an explicit fidelity score per channel, and a lawful-basis manifest attached to every compiled plan rather than a privacy posture asserted at the company level. What they promise, in their words: Privacy-first planning, buying and measurement across CTV, audio, video, display and DOOH. WHERE THEY ARE GENUINELY AHEAD: - No media markup and no platform fee, which is a better commercial structure than a 10% take - Real UK geographic scale and an established audience explorer - Holdout-based measurement built in - A privacy-first architecture with a track record behind it What is actually left for AdBuyMCP: - UK addressable television and cinema, which they do not sell - A published fidelity score and grade on every compiled audience - A lawful-basis manifest per segment rather than a company-level posture - A causal engine that refuses underpowered tests and says why Choose Blis when: Your channel set is the one they cover, UK geographic targeting is the core of the plan, and a structure with no platform fee matters more than cinema and addressable television. For a lot of UK buyers that is the right trade. Choose AdBuyMCP when: Your plan has to include cinema or addressable television, or you want the approximation in every channel's targeting published rather than implied. Sources, all from their own published material: - Blis technology: https://blis.com/our-technology/ ### AdBuyMCP vs AdQuick and Blindspot — https://adbuymcp.com/compare/ooh-specialists Verdict: Better out-of-home tooling than ours. Buy from them if out-of-home is the campaign. AdQuick and Blindspot are the reference products for buying out-of-home properly: exact inventory, visible prices, map-first browsing and rapid launch. AdQuick lists 500,000-plus placements with 110,000-plus audience segments and a public developer API. Both have better screen-level inventory experiences than we do today, and this page is not going to argue otherwise. Two things worth knowing. AdQuick took a $20m strategic investment from OUTFRONT in 2024 that included a three-year exclusive on its sales technology, which is a fair neutrality question to ask of a marketplace. And neither product plans out-of-home against the same audience as your television, audio and cinema, which is the main thing we do that they do not. What they promise, in their words: Plan, buy and measure out-of-home with exact inventory, visible prices and fast launch. WHERE THEY ARE GENUINELY AHEAD: - Screen-level inventory browsing and pricing, done properly - Scale of placements and audience segments - A public API, in AdQuick's case - Speed from brief to live on a single channel What is actually left for AdBuyMCP: - One audience definition and one outcome model across seven channels - Cross-channel frequency deduplication in a single exposure record - UK cinema and addressable television - A neutral position: AdQuick's OUTFRONT investment carries a sales-technology exclusive Choose OOH specialists when: Out-of-home is the campaign rather than a line in one. Their inventory tooling is better than ours and buying a specialist product for a specialist job is usually right. Choose AdBuyMCP when: Out-of-home is one of several channels and you need to know how many people you reached once across all of them. Sources, all from their own published material: - AdQuick products: https://www.adquick.com/products - Why OUTFRONT invested in AdQuick: https://www.adexchanger.com/platforms/why-outfront-invested-20-million-in-adquick-hint-its-about-more-than-money/ - Blindspot: https://seeblindspot.com/book-a-billboard/ ### AdBuyMCP vs Basis Compass — https://adbuymcp.com/compare/basis-compass Verdict: The same idea, built for agencies and for digital channels. The overlap is real and the gap is UK paperwork. Basis Technologies launched Compass in April 2026: agentic planning that takes a brief to a strategy to a media plan to activation across programmatic, direct, search and social. That is the closest thing in the market to what we do, and pretending otherwise would be silly. Three differences. Compass plans across digital channels — there is no out-of-home, cinema or podcast persona unification in it, which is precisely the awkward half of a UK plan. It is agency-oriented rather than brand-operated, with an enterprise sales cycle to match. And it is US-centric. Basis is also a mature company with real scale, which we are not. What they promise, in their words: Agentic planning from brief to strategy to media plan to direct activation, across channels. WHERE THEY ARE GENUINELY AHEAD: - A mature omnichannel platform with agency workflow, billing and reconciliation built in - Real customers, real scale and an established support organisation - Deep programmatic and direct-buy integration - Agency tooling we deliberately do not have What is actually left for AdBuyMCP: - Out-of-home, cinema and podcast unified against the same audience - UK-first supply and UK clearance workflows - Built for a brand to operate directly rather than through an agency seat - A published fidelity score and lawful-basis manifest per compiled channel Choose Basis when: You are an agency, your channel set is digital, and you need planning inside a platform that also handles billing, reconciliation and client workflow. AdBuyMCP has no agency edition and says so. Choose AdBuyMCP when: You are a brand rather than an agency, and your plan has to include the UK channels that transact as paperwork. Sources, all from their own published material: - Basis Compass launch: https://basis.com/news/agentic-ai-solution-for-faster-omnichannel-media-strategy-planning-and-activation ### AdBuyMCP vs Caasie — https://adbuymcp.com/compare/caasie Verdict: They proved the pattern. If you want DOOH with no minimum and no conversation, use them. Caasie is a self-serve digital out-of-home platform with map-based browsing across Australia, the United States, Canada and six UK cities, bought pay-per-play with no contracts and no minimum spend. It is bootstrapped, with a four-person team and revenue around $215,800 in 2024. It is not a scaled competitor and this page will not inflate it into one. It is on this list because it got something right that larger platforms did not: no minimums, map-first browsing, and a price you can see before you commit. Its own commercial structure is worth noting honestly — no platform fee, with the margin embedded in the play price. That is not an absence of a fee, only an undisclosed one. What they promise, in their words: Self-serve digital out-of-home with no contracts, no minimum spend and no platform fees. WHERE THEY ARE GENUINELY AHEAD: - No minimum spend at all, and no conversation required - Map-first browsing that is genuinely better than a list of sites - Live in six UK cities alongside Australia, the US and Canada - Simplicity: one channel, done directly What is actually left for AdBuyMCP: - Six channels they do not sell - AI planning and audience compilation - Measurement beyond plays and impressions - A disclosed fee rather than a margin inside the play price Choose Caasie when: You want digital out-of-home, you want it today, and you do not want to talk to anyone. For a small single-channel out-of-home buy they remove more friction than we do. Choose AdBuyMCP when: Out-of-home is one channel in a plan, or you need to know what the media cost separately from what you were charged. Sources, all from their own published material: - Caasie: https://caasie.co/ ## 13. Guides — https://adbuymcp.com/guides 16 guides to how UK paid media actually works. Each one is reproduced here as its opening argument and its questions; the full text is on the page named beside it. ### How UK addressable TV targeting actually works — https://adbuymcp.com/guides/how-uk-addressable-tv-targeting-works You are buying households whose profile over-indexes on your audience. That is worth knowing before you write the brief. UK addressable television lets you buy a household rather than a programme: the ad is swapped in at the set-top box or in the stream, so two homes watching the same show see different advertising. What you select is an intersection of household attributes — Experian Mosaic groups, occupation and affluence attributes — with postcode districts and broadcast regions. What you cannot select is a person, a job title or an intent signal, because a television set has no idea who is in front of it. AdBuyMCP scores the translation from a business audience into that vocabulary at 55 out of 100 and grades it household-level, and no amount of audience detail moves that score. The one line of a brief that does move it is geography, and only downward: an unmappable city costs fifteen points. The practical consequences are a £3,000 minimum burst on Sky AdSmart, a lead time set by Clearcast rather than by the media owner, and a delivery report that arrives as a statement rather than as an API response. Q: Can addressable TV target my customer list? A: Not through AdBuyMCP, and not in the way a digital platform means it. Sky AdSmart supports advertiser-matched audiences through the sales house under its own data terms, but AdBuyMCP has no account-list or CRM ingestion at all, so what it compiles is a semantic audience — sector, seniority, affluence, geography — into household attributes. If matching a named list is the whole requirement, that is a conversation with the sales house rather than a plan we can build today. Q: Why is addressable TV cheaper per thousand than a spot but harder to buy? A: Because you are buying a smaller, selected universe rather than a share of a mass audience, and because the selection has to be built by the sales house against attributes it holds. The cost of that is process: a minimum burst rather than a minimum weight, a booking made by a person, clearance before scheduling, and a delivery statement rather than a report you can pull. The trade is real, and it is why a fortnight of lead time is the number to plan against rather than the media availability. Q: Does a household attribute mean everyone in the house is my audience? A: No, and this is the single most common misreading of an addressable plan. An attribute says the household over-indexes on something — occupation, income, life stage — against the general population. The person in front of the television at the moment your ad plays may be anyone in the house, which is exactly why the fidelity score is 55 and graded household rather than person, and why a B2B brief on this channel carries a warning saying so on the plan itself. ### What a fidelity score is, and why anyone should publish one — https://adbuymcp.com/guides/what-a-fidelity-score-is One number, one question: how much of the audience you described survived the translation. A fidelity score is a number out of 100 attached to a compiled channel plan, answering one question: how much of your audience definition survived the translation into the targeting vocabulary this channel actually accepts. It is not a quality score, not a forecast and not a prediction of performance. On AdBuyMCP the scores run from 85 on CTV, where consent-chained business segments can carry a person-level claim, down to 40 on cinema, where the buy is a projection from venue catchments. Each score carries a grade naming what is really being addressed — person, household, content, geo-cohort or venue — and a written rationale from the compiler that produced it. Publishing the bad ones alongside the good ones is the entire point: a buyer who has been told by four vendors that a billboard can target IT decision makers already suspects it is not true, and the number is the only version of that conversation that leaves them better informed. Q: Is a fidelity score an industry standard? A: No. It is this platform's own construct, computed by its own compilers, and there is no external body defining or auditing it. What makes it useful is not authority but consistency: the same function produces it for every channel and every brief, the rationale is published beside the number, and the scores are comparable across the seven because they answer the same question. Treat it as a disclosure rather than as a certification. Q: Could you not just raise the scores? A: Mechanically yes, which is precisely why the numbers are traceable to code rather than to a marketing decision. The scores are literals in the channel compilers, and the ones that hurt — 40 on cinema, 50 on out-of-home, 55 on addressable TV — are the ones that make the 85 on CTV worth anything. A site that publishes only the flattering number has told a reader nothing about how the rest were arrived at. Q: Does a higher fidelity score mean better performance? A: No, and conflating the two is the misreading most likely to cost money. Fidelity measures how close the targeting is to your description of the audience. Performance depends on the creative, the offer, the medium's attention profile and the competitive context, none of which the score touches. Cinema scores lowest and can outperform a precisely targeted display campaign on the things cinema is good at. ### Designing a geo-lift test that can actually be powered — https://adbuymcp.com/guides/designing-a-geo-lift-test The design decides the answer. Run the power calculation before the flight, not after it. A geo-lift test measures incrementality by holding out matched markets: you buy some places and deliberately do not buy others, and the difference between them after the campaign starts, adjusted for how they differed before it, is the lift. It is the one causal design in media that works without a login, a pixel or a match rate, which is why it suits out-of-home, cinema and television as well as it suits digital. It is also the design most often run in a state where it could never have detected anything, because the sample-size arithmetic was not done first. The number of days a test needs rises with the square of the baseline volatility and falls with the square of the lift you are trying to detect, so a thin, noisy conversion series cannot be rescued by running longer at the margin. AdBuyMCP runs that calculation before the test and returns the required sample days instead of a lift figure when the design cannot support one. Q: How many markets do I need for a geo-lift test? A: More than one on each side, and enough total volume that the exposed series can move against its own daily noise. Our matcher takes the top three controls per test market, and the candidate list is twelve UK conurbations. The number that actually decides feasibility is not the market count but the required sample days, which comes out of the power calculation on your baseline mean and standard deviation. Two exposed markets with high volume and well-matched controls beat six thin ones. Q: Can I run a geo-lift test on a national campaign? A: Only if you are willing to withhold media somewhere, which is what makes it a test rather than a report. A campaign that buys everywhere has no holdout, and the honest answer is that its incrementality cannot be measured this way. The alternative designs — modelled attribution, marketing-mix modelling — produce estimates rather than causal evidence, and should carry a different label when they do. Q: Why refuse to report a number when the test is underpowered? A: Because the number is not a weak result, it is a random one. An underpowered design produces estimates that swing across a wide range from noise alone, so the figure it returns says more about which fortnight you happened to run than about the media. Once that figure is in a slide it is indistinguishable from a measured one, and it will be defended in a budget meeting six months later. Returning the required sample days instead tells you what would have to change. ### How UK cinema advertising is really booked — https://adbuymcp.com/guides/how-uk-cinema-advertising-is-booked Two sales houses, no buying API, and a calendar that starts on a Friday. UK cinema advertising is sold by two sales houses — Digital Cinema Media, which our own research put at around 85% of the market, and Pearl & Dean — and neither exposes a public buying API. You buy an audience guarantee package rather than a slot: a pack of films guaranteeing admissions against a demographic, optionally narrowed to particular titles, venues or postcode catchments, and optionally upgraded to a premium position closer to the feature. The cinema week starts on a Friday because the film schedule does, flights are usually sold in whole weeks, and the two steps that decide your live date are clearance by the Cinema Advertising Association and mastering the ad into a digital cinema package. AdBuyMCP compiles a cinema plan and generates the booking brief with the packs, the venue list and the flight, and grades the targeting 40 out of 100 at venue level, which is what a projection from a catchment is honestly worth. Q: Can I buy cinema advertising for one town? A: Yes, and local buying is the part of cinema most accessible to a small advertiser. Both sales houses sell single-site and local packages, and the practical floor is set by production and clearance rather than by the media: a thirty-second ad has to be made, cleared and mastered whatever the size of the buy. AdBuyMCP enforces a plan floor of £500 a week on cinema lines, which is roughly the level at which the whole exercise is worth doing. Q: How far in advance do I have to book cinema? A: Plan on about eighteen days from approved plan to on-screen, which is the lead time AdBuyMCP's cinema workflow uses, and add production time if the ad does not exist. The two steps inside that window that are outside anyone's control are the CAA clearance and the digital cinema package. Big titles and seasonal packs sell out earlier than the lead time implies, so a campaign built around a specific release needs longer. Q: Does cinema advertising involve personal data? A: The targeting does not. A cinema specification is audience packs, genre packs and venue catchments, all of them aggregate, which is why this is the one channel whose lawful-basis manifest is trivially short and why it sits outside UK GDPR scope for delivery. Your own measurement of the campaign may involve personal data, and that is covered separately under your own basis. ### Buying UK DOOH without a movement-data licence — https://adbuymcp.com/guides/buying-uk-dooh-without-movement-data A billboard cannot know who you are. It can know who tends to walk past it, and when. Digital out-of-home is sold with audience language that mostly rests on mobility data: a licensed panel of location signals used to say which screens over-index on a given segment. If you do not hold that licence — and we do not — you can still build a defensible out-of-home plan from three things that are entirely aggregate: the venue type of each screen, the geographic polygons you want to be seen in, and the hours you want to be seen at. That is a claim about a place and a time rather than about a person, which is exactly how it should be described. AdBuyMCP grades out-of-home 50 out of 100 at geo-cohort level, fixed regardless of how well the brief matched, and its compiler returns an empty screen-ranking list rather than an invented one, because the ranking is the part that needs the licence. Q: Can DOOH really target business decision makers? A: It can buy screens in places where business decision makers are disproportionately present, at hours when they are there. That is a statement about a place and a time, and it is a legitimate way to reach a business audience. It is not targeting a person, and any proposal that describes it as though a screen knows who is in front of it is describing something that does not exist. AdBuyMCP grades the whole channel 50 out of 100 at geo-cohort level for exactly this reason, and the score does not move. Q: Do I need a movement-data licence to buy out-of-home? A: No. You need it to make audience-index claims about specific screens. Without it you can still select by venue type, by polygon and by daypart, which covers most of what a plan actually needs, and you can still buy through the exchanges. What changes is the language on the plan: you describe the place and the hours rather than an audience percentage, and the claim you make afterwards is correspondingly smaller and correspondingly defensible. Q: Why can't a compiled persona activate a DOOH campaign end to end? A: Because the vendor APIs target specific numeric unit identifiers and there is no verified translation from venue types and postcode polygons into those identifiers. Producing one would mean matching a semantic venue class against a vendor's own inventory taxonomy and trusting the result enough to spend money on it. Until that translation is contract-tested against a live catalogue, activation requires hand-picked units, and this site describes the rail as built rather than as live. ### Running paid media from an AI agent without losing control — https://adbuymcp.com/guides/running-paid-media-from-an-ai-agent The agent proposes. A named human holds the money. Everything else is implementation detail. Pointing an AI agent at a paid media account is not risky because the model might be wrong about a headline. It is risky because the interfaces underneath it were built for a person who can read a confusing response and stop. The four controls that actually matter are: the agent uses the same routes, tenant boundary and approval records as the browser, so there is no weaker second path; its authority is a revocable, versioned mandate with an explicit money ceiling and an expiry rather than an API key; consequential actions require a single-use approval bound to the exact bytes a human approved, and fail closed when the request does not match; and an ambiguous supplier response stops the run and waits for a person, rather than being retried. On AdBuyMCP eleven consequential operations have no agent tool at all, which is a stronger guarantee than a permission the agent is asked not to use. Q: Can an AI agent spend money without a human approving it? A: Not on this platform, and the block is structural rather than a setting. Consequential calls fail closed unless the incoming request matches an approval a human granted for those exact bytes; spend runs against a mandate carrying an explicit money ceiling and an expiry; and enabling spend on a live line is a separately granted action that an approval alone cannot perform. Agents also cannot revoke or replace authority, close missions, or resolve approvals — those operations have no agent tool at all. Q: What actually goes wrong when an agent runs advertising? A: The costly failures are operational rather than editorial. The common one is a retried activation after an ambiguous supplier response, which produces two live campaigns spending in parallel. The others are an agent inheriting a human's session and approving its own work, an authority that never expires and is forgotten, and an approval granted for one action being spent on a different one. All four are prevented by design decisions rather than by prompting, which is why the questions worth asking of a vendor are infrastructural. Q: Do I need to trust the model to use an agent safely? A: Less than you would expect, and that is the point of putting the controls in the product rather than in the prompt. The model extracts semantics, drafts creative, explains plans and proposes optimisations. It does not compute budgets — deterministic code does the arithmetic against adapter forecasts — and it never emits a vendor segment identifier, because registry tables do that. Where its output has consequences, a human approval bound to the exact action stands between the proposal and the money. ### What a DSP is, and whether you need one — https://adbuymcp.com/guides/what-is-a-dsp-and-do-you-need-one The fee is the easy part to compare. The harder question is whether you have anyone to drive it. A demand-side platform is software that buys advertising across many publishers through one interface, bidding on individual impressions in real time, and charging a percentage of what you spend through it. Observed platform fees vary more than most buyers expect: around 20% at The Trade Desk on published analysis, roughly 10% to 15% on DV360, and 5% to 15% across the generic tier, on desk research dated 3 August 2026. The fee is the easy thing to compare. The harder question is whether you have somebody whose job is to operate it, because a DSP is a trading interface built for traders, and an unattended seat is the most expensive way to buy media badly. Q: Is a DSP cheaper than an agency? A: On fee, usually. A DSP platform fee of 5% to about 20% against a UK agency's 10% to 30% with a £500 to £1,000 monthly minimum looks favourable, and often is. On total cost it depends entirely on who operates the seat: an agency fee includes a person, and a DSP fee does not. Compare fee plus operator against fee plus service, not fee against fee. Q: Do I need a DSP to buy CTV? A: No. Connected TV can be bought through a DSP, through a specialist performance-CTV platform, directly from some sellers, or through a layer that buys on your behalf. A DSP makes sense for CTV when it is part of a wider programmatic plan you are already running; it makes less sense when CTV is the only programmatic line you have. ### Why most incrementality tests are invalid — https://adbuymcp.com/guides/why-most-incrementality-tests-are-invalid An underpowered test does not give a weak answer. It gives a random one. Most incrementality tests fail before the media runs, and for two reasons that are both checkable in advance. The first is power: if the design cannot detect an effect of a plausible size given the spend, the conversion volume and the variance in the data, then any number it produces is noise that happens to look like a result. The second is contamination: if the control markets received some of the campaign, the comparison is exposed against slightly-less-exposed and the measured lift is an underestimate of unknown size. Neither failure announces itself in the output — both produce a confident-looking percentage — which is why the checks have to happen before the money is committed rather than after. Q: Can I run an incrementality test on a small budget? A: Sometimes, and the deciding factor is rarely the budget alone. It is the combination of conversion volume, baseline variance and the number of independent markets you can separate. A multi-catchment retailer with twenty locations and steady weekly bookings can often power a test on a modest budget; a national brand with one campaign and lumpy conversions frequently cannot power one at any budget. The power calculation answers it in advance, which is the point of running it first. Q: Is marketing mix modelling a substitute? A: No, though it answers an adjacent question. Modelling estimates each channel's contribution from historical patterns without needing identifiers, which makes it valuable for channels that cannot be tracked. It remains correlational, hungry for history and sensitive to specification — two competent modellers can produce materially different answers from the same data. It is evidence, but it is not the same evidence as a controlled experiment, and presenting a model output as proof of causation is the failure the four-verb model exists to prevent. ### Frequency capping across channels, and why it usually is not happening — https://adbuymcp.com/guides/frequency-capping-across-channels Seven platforms each capping at four is not a cap of four. Frequency capping limits how many times one person or household sees the same advertising in a period. Nearly every platform does it within its own inventory, and nearly none does it across sellers, which means a plan running on seven channels with a cap of four on each is not capped at four — it is capped at up to twenty-eight, against households that appear in several of the seven. Doing it properly requires one exposure record that every channel writes into, keyed on something all of them can supply. That is a structural requirement rather than a setting, and it is the reason cross-channel frequency is claimed far more often than it is delivered. Q: Can I just set the same cap on every platform? A: You can, and it does not produce the cap you set. Each platform enforces it within its own inventory only, so the aggregate exposure for a household appearing on several channels is the sum rather than the maximum. Setting four on seven channels permits up to twenty-eight. There is no configuration that fixes this from inside the platforms, because none of them can see the others. ### What Clearcast and the CAA actually check — https://adbuymcp.com/guides/what-clearcast-and-the-caa-actually-check The clearance body is not judging your creative. It is asking for your evidence. UK broadcast and cinema advertising must be approved before it runs, by separate bodies applying separate rules, and the approval is required rather than advisory. Three provisions account for most rejections. Objective and superlative claims — best, number one, market-leading — need documentary substantiation held by the advertiser. Medicines, medical devices and health claims require evidence and unsubstantiated efficacy claims are routinely refused. And cinema copy containing strong language attracts an age rating that restricts which films it can be shown beside, which is a media constraint as much as a creative one. None of these is a judgement about quality; all of them are requests for evidence or consequences of category. Q: How long does clearance take? A: Long enough to be the binding constraint rather than the media availability. For cinema, budget around eighteen days from brief to on-screen once clearance is in the path. Broadcast varies with the complexity of the claims: straightforward copy moves quickly, and anything requiring substantiation review takes as long as the review does. Neither is a process to start after the flight has been booked. ### MCP for marketers: what the protocol actually changes — https://adbuymcp.com/guides/mcp-for-marketers The protocol is a door. What matters is what is behind it. The Model Context Protocol is an open standard that lets an AI assistant call external tools and data through one consistent interface, so a model can operate software rather than only describe it. For a marketer the practical change is real: you can ask an assistant to compile an audience, produce a plan and pull a report, in conversation, instead of clicking through a dashboard. What it is not is a capability in itself. MCP is an integration standard — several advertising platforms already expose one — and a vendor leading with the protocol as a differentiator is describing the door rather than the building. The question worth asking is what sits behind it, and specifically what an agent is prevented from doing. Q: Is an AI agent safe to point at a media budget? A: It depends entirely on what stands between the agent and the money, and that is an architectural question rather than a question about the model. The things to establish: that the agent uses the same execution path as a human rather than a parallel one, that consequential operations are absent from its surface rather than merely discouraged, that authority carries an explicit ceiling and expiry, and that an ambiguous supplier response stops the run rather than being retried. Where those hold, the model's fallibility is contained by design rather than by prompting. ### UK media buying for B2B marketers — https://adbuymcp.com/guides/uk-media-buying-for-b2b-marketers The committee is six to ten people. You are reaching two of them. A considered B2B purchase, and not only a large one, is decided by a buying committee of roughly six to ten people, and most B2B media plans reach the two or three who are professionally active on LinkedIn during working hours. Reaching the rest means leaving the platforms that know job titles, which means accepting a proxy: consent-chained business segments on connected TV, household attribute combinations on addressable television, venue and daypart inference on out-of-home. Those proxies are genuinely weaker, and the useful skill is knowing how much weaker. Connected TV can carry a person-level claim at 85 out of 100 where business segments match strongly; a billboard is a geo-cohort at 50 and cannot be anything better. Q: Can I target job titles on CTV? A: Not directly, and nobody can. What you can buy are consent-chained business segments built from job-title and intent data and delivered against streaming households. AdBuyMCP grades that person-level and scores it 85 out of 100 when the match is strong, which is the honest description of a very good proxy rather than of an exact match. A television has no idea who is in front of it. Q: Is a household-level proxy good enough for B2B? A: It depends on your contract value and your patience. At high contract values the waste is affordable and the incremental reach is the point; at low ones it is not. The useful discipline is to decide in advance what evidence would justify a second campaign, because a B2B surround strategy judged on last-click after four weeks will always look like a failure regardless of whether it worked. ### How to read a media plan — https://adbuymcp.com/guides/how-to-read-a-media-plan The interesting fields are the ones most plans leave out. A media plan is a budget split with forecasts attached, and the quality of one is decided by how much it discloses rather than how it looks. The fields worth checking are: which supplier actually executes each line, what the forecast reach, frequency and CPM are and what they were derived from, what the data segments cost on top of the media, what the targeting approximation is worth, and why each line is there at all. Most plans disclose the first and the last poorly and the middle three not at all. The absence is rarely deliberate concealment — it is more often that the planner does not have the numbers either — but the effect on you is identical. Q: What is the single most useful question to ask about a media plan? A: What does each line cost me in total, including data and fees, and who executes it? Those two together surface most of what is usually hidden: the segment uplift that sits on top of the media, the intermediary chain between you and the publisher, and whether a line is a programmatic buy or a piece of paperwork with a lead time. A plan that answers both readily is a plan whose author has done the work. ### First-party data in UK media buying — https://adbuymcp.com/guides/first-party-data-in-uk-media-buying The bridge from your database to a television exists. Most of it is not built yet. First-party data is what you collected from your own customers, and it is both the most valuable targeting asset most advertisers own and the hardest to activate on premium channels. Using it on social platforms is straightforward because they hold matching identifiers. Using it on connected television, audio or out-of-home requires an identity bridge that does not compromise the list, which in the UK and Europe means a consent-based identifier derived from a hashed email or phone number. Those standards exist and are the right architecture. Being honest about the current state matters more than the architecture: this route requires an operator agreement, and AdBuyMCP's compiled plans carry no such identifier today. Q: Can I upload a customer list to target on CTV? A: Not through AdBuyMCP today. The consent-based identifier that would carry a first-party list onto streaming inventory requires an operator agreement, and the adapter is an explicit non-transacting stub until one exists — the compiler emits an empty identifier list, so no plan carries one. On the social platforms, list matching works because they hold the identifiers themselves, and that route is unaffected. ### Choosing between cinema and CTV — https://adbuymcp.com/guides/choosing-between-cinema-and-ctv They are not the same buy, and the deciding factor is rarely the price. Cinema and connected television look adjacent — both are full-screen video with sound, bought against an audience — and behave completely differently. Connected TV targets households, can carry a person-level claim at 85 out of 100 where business segments match, launches in days and reports impressions. Cinema targets a venue and a film, scores 40 out of 100 as a venue projection, takes around eighteen days once clearance is in the path, and delivers an attention profile nothing else in a media plan approaches. The decision is rarely about price. It is about whether you need precision and speed, or attention and impact, and about whether the campaign has a date it must hit. Q: Is cinema worth buying at a small budget? A: For a launch, often. For ongoing performance, usually not. Cinema is bought in weekly bursts against a venue list and rewards concentration — the same money spread thinly across a region produces too few admissions in any one catchment to register. The plan floor here is £500 a week for that reason, and a two-week regional burst is the smallest shape that does anything. ### Sky AdSmart explained — https://adbuymcp.com/guides/sky-adsmart-explained Household attributes, postcode districts, and a booking process rather than an auction. Sky AdSmart is addressable television: two households watching the same programme at the same time receive different adverts, swapped in at the set-top box. What you select is an intersection of household attributes — Experian Mosaic groups, occupation and affluence attributes, life-stage indicators — with postcode districts and broadcast regions. Entry campaigns have historically started around £3,000 per burst through resellers. It is booked rather than bought: there is no self-serve auction, the order goes through a sales house, and clearance sits in front of the flight. The targeting is genuinely powerful for a geographic or demographic audience and is a proxy at 55 out of 100 for a business one. Q: Can I target a job title on Sky AdSmart? A: No. You can select household attributes that model occupation and seniority — business owners and directors, senior decision makers in business — which are modelled household characteristics rather than verified titles. A set-top box has no idea who is watching. That is why the honest fidelity grade for a business audience here is household at 55 out of 100, and why the score does not improve with a better brief. ## 12b. What UK media costs — https://adbuymcp.com/cost What each of the seven channels actually costs in the UK: the market minimum, the floor we enforce when we build a plan, and everything else that leaves your account. The two kinds of number are labelled, because one is a product fact and the other is dated desk research. Two kinds of number appear below and they are not interchangeable. The enforced plan floor is a product fact: a plan line that cannot clear it is dropped and its budget redistributed. The market minimum is desk research dated 3 August 2026 describing what other sellers ask for, and it is not a commitment by anyone. Below about £10,000 a month in total media, this is the wrong product. A 10% fee on £2,000 is £200 against the buyer who needs the most support from us, and a multi-channel plan at that budget is too thin to pace on any line and far too thin to power a causal test. Buy search and social directly, and come back when there is a second channel worth funding. ### CTV advertising cost — https://adbuymcp.com/cost/ctv-advertising-cost Connected TV is bought on a cost per thousand impressions, and self-serve minimums have collapsed over the last two years — the published reference points now sit around fifty dollars a day at the low end, with no campaign minimum at all on some platforms and observed CPMs in the range of fifteen to thirty dollars. Those are dollar figures because they are the closest published comparators: the UK sellers do not publish rate cards. AdBuyMCP enforces its own floor of £140 a week on a CTV line, below which the line is dropped from the plan and its budget redistributed. The number that surprises people is not the media: it is the data. A consent-chained business segment adds roughly £2.50 to £3.00 per thousand impressions on top of the media cost, which on a targeted B2B buy can be a fifth of the CPM again. Enforced plan floor: £140 per week Market minimum (research, Aug 2026): Around $50 a day at the low end, with some platforms carrying no campaign minimum and others setting programme minimums of $10,000 to $25,000. Observed CPMs of roughly $15 to $30. How it is priced: A CPM, set at auction rather than from a rate card, so the figure on a plan is a forecast rather than a price. What moves it: how narrow the audience is, how much competition there is for the same households, and whether you are buying premium streaming inventory or the long tail. A tightly targeted B2B buy in London will clear well above a broad national one. Smallest sensible buy: At the £140 weekly floor you are buying a few thousand impressions against a narrow audience, which is enough to check that trafficking and creative work end to end and not enough to learn anything about performance. A four-week flight at £1,000 to £2,000 a week is the smallest buy from which a reach and frequency figure means anything, and a powered lift test needs considerably more than that. What catches people out: That the data can cost a fifth of the media again. A £15 CPM with a £3 segment uplift is an £18 CPM, and platforms that quote you the first number and invoice the second are relying on you not to check. Other costs: Data segment uplift — Roughly £2.50 per thousand on Bombora intent topics and £3.00 on Dun & Bradstreet segments. It appears on the plan before you approve it rather than in the reconciliation afterwards. Creative — A CTV creative is a mezzanine file plus its trafficking metadata. If you are producing rather than reusing, AI video generation here runs from £6 for ten seconds standard to £15.60 premium, charged separately from the media. The platform fee — 10% of the media, or 5% plus £749 a month above £25,000 of trailing thirty-day spend. ### Addressable TV advertising cost — https://adbuymcp.com/cost/addressable-tv-advertising-cost UK addressable television — Sky AdSmart and its equivalents — has historically started around £3,000 for an entry campaign bought through a reseller, and that burst minimum is the number most first-time buyers meet. AdBuyMCP enforces a £1,000 floor on an addressable line in a plan, which is lower because a plan line is not a campaign: it is one channel's share of a larger budget, and the seller's own minimum still applies at booking. Two things to be clear about before reading any further. This channel grew 37% to £1.84bn in 2025 on AA/WARC figures, so the enthusiasm is warranted. And we do not buy UK broadcast inventory today: these are booking-paperwork rails, and no addressable delivery has ever been reconciled through us. Enforced plan floor: £1,000 per line Market minimum (research, Aug 2026): From around £3,000 for an entry campaign via Sky AdSmart resellers. Comcast's Universal Ads launched UK self-serve in 2026 with low or no minimum, which is actively reshaping this number. How it is priced: A CPM against households selected on attributes, with the price rising as the selection narrows. Because a household is bought rather than an impression against a person, the effective cost per person reached depends on household size, which is a variable nobody controls. Smallest sensible buy: A £3,000 to £5,000 burst over two to four weeks in a defined region. Addressable television rewards concentration: the same money spread nationally buys a frequency of roughly one, which is below the threshold at which television does anything at all. What catches people out: That the lead time is the constraint rather than the money. This is booked as paperwork with clearance in front of it, so a campaign that has to be live in ten days usually cannot be. Other costs: Clearance — Copy must be approved before it airs, and superlative or health claims need documentary substantiation. That is time rather than a line item, but a rejected script can cost a fortnight. Production — A broadcast-spec mezzanine at 1920×1080, 25fps, at least 20Mbps, with supers inside title-safe and audio to EBU R128. Reusing a social cut will not clear. The platform fee — 10% of media, or 5% plus £749 a month above £25,000 of trailing 30-day spend. ### Podcast and audio advertising cost — https://adbuymcp.com/cost/podcast-advertising-cost UK digital audio is cheap to start on, though not the cheapest here: paid search carries no media minimum at all and the DOOH plan floor is £100 against audio's £250. Spotify's self-serve ads manager starts around £250 a campaign and Acast's at about $250, on desk research dated 3 August 2026. AdBuyMCP's own plan floor matches that at £250. The number to watch is not the minimum but the spread between formats. A dynamically inserted spot — targeted, reportable, removable — is bought on a CPM at the low end of this market. A host-read endorsement baked permanently into an episode is closer to $5,000 and cannot be turned off once published. They are sold in the same conversation and priced an order of magnitude apart. Enforced plan floor: £250 per line Market minimum (research, Aug 2026): £250 a campaign on Spotify's UK self-serve ads manager; around $250 on Acast self-serve; approximately $5,000 for a host-read spot. How it is priced: Dynamically inserted audio is a CPM, priced on how narrow the targeting is and how premium the inventory. Host-read is priced per episode or per thousand downloads against a specific show, negotiated rather than auctioned, and usually sold in packages. Smallest sensible buy: £1,000 to £2,500 across a four-week flight on genre and show targeting. Audio is a strong response channel at small budgets — a promo code or vanity URL gives you observed response cheaply — and a weak causal one, because a powered lift test usually needs more spend than an audio line carries on its own. What catches people out: That the daypart matters more than the budget. A business audience is reached on the commute; the same money spent across the day reaches a different, larger and less relevant audience. Other costs: Voiceover and production — A script with no read direction is half a brief. A generated UK voiceover here costs £1.25 and a music bed 50p, charged separately; a professional read costs considerably more and is often worth it. Named-show deals — Buying a specific show happens through a private marketplace deal, which is negotiated before it can be executed. That is a lead time rather than a fee. The platform fee — 10% of media, or 5% plus £749 a month above £25,000 of trailing 30-day spend. ### DOOH advertising cost — https://adbuymcp.com/cost/dooh-advertising-cost Digital out-of-home has almost no floor left. Programmatic DOOH through the major supply platforms starts in the low hundreds of pounds, and at least one self-serve platform sells pay-per-play with no minimum spend and no contract at all. AdBuyMCP's own plan floor is £100, the joint lowest of the seven channels. What that low entry price hides is that a small DOOH buy is very easy to waste: a few hundred pounds spread across a city buys a handful of plays on screens nobody specific walks past, and the difference between a cheap DOOH campaign and an effective one is concentration rather than budget. Enforced plan floor: £100 per line Market minimum (research, Aug 2026): No minimum at all on pay-per-play platforms; low hundreds of pounds through programmatic supply platforms. How it is priced: Either a CPM against estimated impressions or a cost per play on the screen. Pay-per-play is the more honest unit for a screen, since an impression estimate on a billboard is a model rather than a count, but it makes cross-channel comparison harder because everything else is priced per thousand. Smallest sensible buy: £1,500 to £3,000 concentrated on one venue type in one catchment over two to four weeks. Concentration is everything here: the same money across a whole city produces a frequency low enough that nobody notices the campaign happened. What catches people out: That seven words is the real constraint. A DOOH headline nobody can read at walking pace is spend rather than creative, which is why the copy limit here is enforced in code rather than advised. Other costs: Creative rendering — Screens run at specific resolutions — 1080×1920, 1920×1080, 1400×400 — and a creative that does not match is rejected. Generating one here is 50p to £2 depending on whether it moves. Audience data — Where a consent-attested movement segment is used it carries its own cost. Worth knowing that we do not currently hold that contract, so our DOOH plans use venue type and polygon rather than movement data. The platform fee — 10% of media, or 5% plus £749 a month above £25,000 of trailing 30-day spend. ### Cinema advertising cost — https://adbuymcp.com/cost/cinema-advertising-cost UK cinema does not have a published price, and that is the most useful thing to know about it. It is sold quote-based through two sales houses — one of which covers around 85% of UK screens — packaged by audience pack, film genre and venue rather than bought at auction. AdBuyMCP's plan floor is £500 a week, which reflects how cinema is actually sold: in weekly bursts against a venue list. Nobody offers true self-serve cinema in the UK, including us, and any platform claiming to be first at it is wrong — the main sales house has run a campaign planner and established buying workflows for years. Enforced plan floor: £500 per week Market minimum (research, Aug 2026): Quote-based. No published self-serve minimum exists; packaged regional bursts are the usual entry point. How it is priced: Per admission or per screen-week against a selected venue and film package, quoted rather than auctioned. Price moves with the certificate and the release: a package against a major opening weekend costs a multiple of the same venues in a quiet week. Smallest sensible buy: A two-week regional burst against a defined venue list. Cinema is bought in bursts because a single week rarely accumulates enough admissions in one catchment to register, and the flight default here is two weeks for that reason. What catches people out: The lead time. Around eighteen days from brief to on-screen once clearance is in the path, which makes cinema impossible to bolt onto a campaign late. Other costs: Print and delivery — A Digital Cinema Package at 2K flat, 24fps, 5.1 audio, sixty seconds maximum, routed for distribution. This is a real production cost and it is not the same as a broadcast master. Clearance — Cinema has its own clearance body and its own age rating, separate from broadcast. Strong language restricts pairing to 15 and 18 certificates, which changes the audience you can reach. The platform fee — 10% of media, or 5% plus £749 a month above £25,000 of trailing 30-day spend. ### Paid search cost — https://adbuymcp.com/cost/paid-search-cost Paid search is priced per click at auction, and the media cost is entirely a function of your category's competitiveness — there is no useful national average and anyone quoting one is guessing. The number this page can actually answer is the management cost. UK agencies charge 10% to 30% of spend, most commonly 15% to 20%, with minimum monthly fees of £500 to £1,000 and typical retainers between £1,250 and £3,500. AdBuyMCP charges 10%, or 5% plus £749 a month above £25,000 of trailing thirty-day spend, with a plan floor of £100. If search is the only thing you buy, the platforms' own tools are free and you should use them. Enforced plan floor: £100 per line Market minimum (research, Aug 2026): No media minimum. Agency management from £500 to £1,000 a month minimum. How it is priced: Cost per click, set at auction against everyone else bidding on the same query. What moves it: how commercially valuable the query is, how relevant your landing page is judged to be, and how many competitors want the same click. Smallest sensible buy: Enough to accumulate clicks in a fortnight, which in a competitive B2B category might be £2,000 and in a niche one £300. The floor here is £100 because a search line inside a multi-channel plan can be genuinely small and still useful as a demand-capture layer under a brand campaign. What catches people out: That the layer is usually not worth paying for on search alone. This page exists to say so: the argument for AdBuyMCP is deduplicating search against six other channels, not running search better than you already do. Other costs: Management — Whatever you pay someone to run it. The comparison above is the point of this page: a 10% platform fee against a 15% to 20% agency fee with a £500 to £1,000 minimum. Landing pages — The largest hidden cost in search, and not one anybody invoices you for. A campaign pointed at a page that does not convert is spend rather than marketing. ### Paid social cost — https://adbuymcp.com/cost/paid-social-cost Paid social is priced at auction on a CPM or a cost per action, and the price depends on your audience, your creative and your objective in ways no external average captures. It is also the channel where a third party adds least: Meta and TikTok optimise their own inventory using signals nobody outside can see, and their managers are free. AdBuyMCP's plan floor is £250. The honest reason social is here is not better buying — it is that a plan which ignores the channel carrying most of your budget cannot deduplicate your audience or support an incremental-reach claim. Enforced plan floor: £250 per line Market minimum (research, Aug 2026): No meaningful media minimum on either platform. Agency management from £500 to £1,000 a month minimum, or 10% to 30% of spend. How it is priced: CPM or cost per action at auction. The dominant variable is creative: on social, a good creative can cost a fraction per outcome of a poor one against the same audience, which is not equally true of the other six channels. Smallest sensible buy: Enough for the platform's optimisation to leave its learning phase, which usually means a few hundred pounds a week per ad set rather than a token daily budget spread across many. Under-funding many ad sets is the most common way a small social budget is wasted. What catches people out: That buying social through a layer usually costs more than buying it directly and is still sometimes right — because the thing you are paying for is the deduplication against the other six channels, not the social buying itself. Other costs: Creative volume — Social consumes creative faster than any other channel. Generating variants here costs 50p an image and £6 for ten seconds of standard video, charged separately from media. AI provenance on TikTok — AI-generated video must carry a self-disclosure declaration. That is a compliance requirement rather than a cost, and missing it blocks the upload. Observed fee benchmarks across the category, desk research dated 3 August 2026: - AdBuyMCP: 10% of media (5% plus £749 a month above £25,000 trailing 30-day spend) - The Trade Desk: ~20% of spend (20.3% in 2024; 1–2 points negotiated at $500k–750k thresholds) - DV360: ~10–15% (Platform fee) - Generic DSP tier: 5–15% (Technology fee on spend) - UK agencies: 10–30% (Most commonly 15–20%, with £500–£1,000 minimum monthly fees) - MNTN: Undisclosed (Its own documentation states it keeps a percentage and does not disclose the margin) ## 13b. How-to procedures — https://adbuymcp.com/how-to 10 procedures for the things we actually ask you to do — writing a brief that compiles well, reading a fidelity score, running a lift test that can answer, granting an agent authority and taking it back. ### Write a brief an agent can compile — https://adbuymcp.com/how-to/write-a-brief-an-agent-can-compile The quality of a compiled plan is almost entirely a function of the brief, and the scoring makes that visible rather than leaving it to be discovered. A persona label with no product or offer context compiles at 45 out of 100 on paid search; the same audience with buyer-authored intent topics compiles at 82. Neither number is a judgement of your brief — they are the honest distance between what you described and what the channel can be sold. What follows is what to include so that distance is as small as it can be. Before you start: A clear idea of who you want to reach, in your own words rather than in segment names; What you sell and what the specific offer is; A budget and a flight, even approximate; The geography, named as UK cities or postcode districts Steps: 1. Describe the person, not the segment — Write who you want to reach in plain English: sector, seniority, what they do, where they are. The language model extracts semantics from this; it never picks a vendor segment identifier, because deterministic registry tables do that by matching your terms against catalogue keywords. Watch out: Do not write segment names you have seen in a DSP. They will be matched as text and will usually score worse than the plain description would have. 2. Name the geography in mappable terms — Use UK cities or postcode districts. The compiler resolves cities to postcode districts and to BARB television regions through an explicit table. Watch out: A city the geography table cannot map costs fifteen points on every channel, and the rationale names the city. If you need somewhere unusual, say so in the session rather than hoping. 3. Add what you sell and what the offer is — This is the single highest-value addition. It is what moves paid search from a persona-label fallback at 45 to an interest-to-query projection at 68. Watch out: Without it the search plan is explicitly marked as not activation-ready, which is the compiler telling you it has not got enough to work with. 4. Add intent topics if you have them — The actual phrases your buyers type when they are looking for what you sell. These become plain-language seeds that the provider's own keyword planner resolves into ideas with volume and bid data. Watch out: This takes search from 68 to 82. Interests are not intent: 'interested in cybersecurity' and 'searching for endpoint protection pricing' score differently and should. 5. State exclusions — Anything you do not want to reach or appear beside. Exclusions veto overlapping catalogue entries and are reported separately per channel. Watch out: Where a channel cannot express your exclusion natively, the plan says so rather than silently dropping it. Social exclusions in particular are not compiled into audience identifiers. 6. Read the seven fidelity scores before the budget — The compiled plan returns a score and a written rationale per channel. Read the rationales: they say what matched, what did not, and what would improve it. Watch out: If a channel scores lower than you expected, the rationale usually names the missing input rather than a limitation of the channel. Result: A persona that compiles at the top of each channel's available band, with rationales that tell you what any remaining gap is caused by. ### Read a fidelity score — https://adbuymcp.com/how-to/read-a-fidelity-score A fidelity score is a published number out of 100. It says how much of your audience definition survived translation into one channel's native targeting vocabulary, and it is not a forecast, not a quality rating and not a prediction of performance. The most useful thing to know when reading one is whether it can move at all: four of the seven channels return the same constant however you describe the audience, and three vary across three discrete values each. Only the geography line moves the fixed four, and it can only cost them: an unmappable city takes fifteen points off every channel. Trying to improve a fixed one by writing more about the audience is wasted effort. Before you start: A compiled plan, which the sandbox will produce without credentials; Five minutes and a willingness to read the rationale rather than only the number Steps: 1. Read the grade before the number — Grades run person, household, content, geo-cohort and venue, each one step further from an individual. The grade tells you what kind of claim the channel can support; the number tells you how well it did inside that. Watch out: An 85 person-level and a 65 content-level are not comparable on a single scale. They are different claims about different objects. 2. Check whether the score can move — Audio returns 65, addressable TV 55, DOOH 50 and cinema 40, before the geography penalty. CTV returns 85, 65 or 60; search returns 82, 68 or 45; social returns 62, 48 or 38. Watch out: If a fixed channel scores lower than you want, the answer is to spend less there or not buy it, not to rewrite the audience. 3. Read the rationale — Every score comes with a written rationale saying what produced it — which segments matched, at what confidence, and what was used as a fallback. Watch out: A CTV plan graded household at 65 rather than person at 85 means segments matched only at low confidence and the plan was flagged for human review rather than promoted. That is a real distinction and it is in the rationale. 4. Look for the geography penalty — Naming a city the UK geography table cannot map costs fifteen points on every channel, floored at ten, and the rationale names the city. Watch out: This is the most common single cause of an unexpectedly low score across all seven channels at once. If every channel dropped, look here first. 5. Decide what to fund — Use the scores to decide where the money goes rather than to decide whether the plan is good. A 40 on cinema is not a failure; it is the honest worth of a venue projection, and cinema may still be the right buy for a launch. Watch out: The trap is treating the score as a ranking of channels. It is a ranking of how well each one can express your audience, which is a different question from which one will work. Result: A clear view of which channels can express your audience well, which cannot and never will, and which would improve with a better brief. ### Run a geo-lift test — https://adbuymcp.com/how-to/run-a-geo-lift-test A geo-lift test runs advertising in some markets, withholds it from comparable others, and compares the outcomes. It is the causal design that needs no login, pixel or match rate, which makes it the only credible option for out-of-home, cinema and television. Most such tests fail before they start, for two reasons: not enough power to detect a plausible effect, or a control group that received some of the campaign. Both failures are steps in this procedure rather than errors in it, and the engine returns each one with the specific thing that would fix it. Before you start: Enough geographically separable markets — a national campaign in one market cannot be tested this way; An outcome you can observe by geography: bookings, leads, point-of-sale, or conversions with a location; Pre-campaign history in those markets, so they can be matched on behaviour rather than on demography; Enough spend that a plausible effect would be detectable, which is what the power step establishes Steps: 1. Nominate a test market — Choose where the media will run. AdBuyMCP matches candidate control markets against it on the pre-campaign outcome series rather than on demographic similarity. Watch out: Matching on demography is the intuitive choice and the wrong one. Two cities that look alike on paper often behave differently commercially, and it is the behaviour you need to match. 2. Let the power analysis run first — A two-sample calculation on daily observations runs before any readout, defaulting to 5% significance and 80% power. It returns either a powered verdict with the minimum detectable lift, or an underpowered one. Watch out: This is the step that saves the money. An underpowered test does not produce a weak answer, it produces a random one. 3. If it refuses, read what it needs — An underpowered design returns the number of days of post-exposure data required and the minimum lift the design could detect, with no lift figure at all. Its own words are that reporting a number from it would be reporting noise. Watch out: If the minimum detectable lift is 30% and a realistic lift is 5%, the test as designed cannot answer your question. The engine returns the sample days that would close the gap instead of a number, and if that is longer than the flight you can fund, the test should not be commissioned. 4. Check the control is genuinely held out — If every candidate control market received delivery from the campaign, none of them is a holdout and the engine refuses on that basis, naming the contaminated markets. Watch out: A contaminated control is the most common way a lift test quietly measures nothing, because it still produces a confident-looking number. Exclude those markets from targeting for the test window, or nominate a control the campaign does not buy. 5. Run the flight without touching the design — Let the campaign run. Changing the targeting, the budget split or the market list mid-flight invalidates the comparison. Watch out: The temptation to optimise into the control markets when the test markets are performing is strong and fatal. 6. Read the difference-in-differences result — A powered design reports the lift with its confidence interval, comparing the change in exposed markets against the change in control markets so that anything affecting both cancels out. Watch out: A result is graded causal only when every geo outcome is observed, every delivery row maps to a live line, and control exposure is verified. Sandbox delivery is labelled modelled by construction, so a sandbox campaign can never reach a causal grade. Result: Either a lift figure with a confidence interval that can survive being questioned, or a documented refusal naming exactly what would have to change. ### Check a lawful-basis manifest — https://adbuymcp.com/how-to/check-a-lawful-basis-manifest Every compiled channel plan carries a lawful-basis manifest: each data source used, the basis it relies on, the conditions attached to that basis, and any warnings raised while compiling. It exists so the answer to "where did this audience come from?" is a document rather than a conversation. The most important thing to understand before reading one is what it is not. It is an engineering artefact that records what remains to be done, not a compliance certificate — and the conditions in it are obligations on you. Before you start: A compiled plan for at least one channel; Knowing who in your organisation owns data protection decisions, because some conditions are theirs to satisfy Steps: 1. Check which channels are in scope at all — Cinema and DOOH delivery involve no personal data — venue catchments, OpenOOH venue types, postcode polygons and dayparts — and the compiler records them as outside UK GDPR scope rather than leaving it to be inferred. Movement data is not in that list because none reaches the product: the Adsquare contract does not exist, so the screen-ranking list the compiler emits is empty. Watch out: Outside scope describes the delivery, not everything used to choose it. Cinema's manifest really is outside scope throughout, but a DOOH plan that references a consent-attested segment still carries that segment's basis, and the manifest names the provider that attested it. Read it rather than assuming both channels are clear. 2. Read the basis per source, not per campaign — Different segments in the same plan can rely on different bases. The manifest records one per source: consent-chained, legitimate interests, or outside scope. Watch out: Per-campaign is the wrong granularity and is how a single non-compliant source hides inside an otherwise clean plan. 3. Read the conditions as a to-do list — A consent-chained source attaches a condition that the supply chain must be framework-validated and the vendor attestation verified on connection. A legitimate-interests source attaches a condition that an assessment must be on file. Watch out: These are things you owe, not things already done. A manifest listing a condition is telling you the work is outstanding. 4. Read the warnings — Low-confidence segment matches, unmapped geographies and excluded segments that could not be expressed natively all raise warnings that stay attached to the compiled plan. Watch out: A warning that a persona exclusion could not be expressed on a channel is the one most likely to matter to a brand-safety reviewer, because it means the exclusion is not enforced there. 5. Take it to whoever owns the decision — The manifest is designed to be the thing you hand a DPO rather than a summary you write for them. It names providers, bases, conditions and warnings in one place. Watch out: What it will not do is make a judgement. Whether legitimate interests is appropriate for your processing is an assessment somebody has to make and document. Result: A written record of what made each targeting decision lawful, and an explicit list of the conditions still outstanding on your side. ### Fund a campaign — https://adbuymcp.com/how-to/fund-a-campaign-wallet Funding runs through a prepaid wallet with a £100 minimum top-up. The check at launch covers the media plus the maximum platform fee on it rather than the media alone, so a campaign cannot go live with enough money for the buy and not enough for the fee. If the wallet is short, the refusal names the exact amount rather than declining generically. One gate matters more than all the others: a wallet holding sandbox-minted credit is refused for live execution outright, so a sandbox balance can never become real spend. Before you start: An approved plan, with prechecked creative bound to every line; A payment method, for a live campaign; For live media specifically: a design-partner engagement, because no rail is spend-verified Steps: 1. Top up the wallet — The smallest credit the buyer-facing contract accepts is £100. Top-ups, spend, asset charges, asset refunds and the platform fee are five distinct entry kinds in one ledger. Watch out: That ledger design is why a statement reads as a sequence of events rather than a single total, and why the fee sits beside the media it was charged on. 2. Understand what the launch check reserves — The funding check at launch covers the media and the maximum platform fee on it, under serialisable isolation so two simultaneous launches cannot both pass against the same balance. Watch out: Budgeting for the media alone is the most common reason a first launch is refused. 3. Read the shortfall if it refuses — An insufficient balance returns the exact amount short — "wallet is £X short of the launch commitment" — rather than a generic decline. Watch out: The check fails closed by default. Only a sandbox preview may opt into warning instead, which is the right way round: a preview that warns is useful, a live buy that warns is a bill. 4. Know that asset charges are a harder gate — AI asset production has no warn-only path at all, because a real provider cost is incurred the moment a job is submitted. If the wallet cannot cover it, it does not submit. Watch out: A definite provider failure refunds automatically. An ambiguous outcome retains the charge and disables retry, so a duplicate render cannot be produced and billed, and it is cleared by a named human. 5. Expect funds to stay held through reconciliation — On a live line the reservation is not released until spend is reconciled to the provider's own invoice or account export, and the final match is rechecked in the same transaction as the release. Watch out: Missing, mismatched or stale evidence keeps the money held. That is deliberate: releasing on an unverified report is how a reconciliation gap becomes a write-off. Result: A funded campaign with the fee reserved alongside the media, or a refusal naming the exact shortfall. ### Install the MCP server — https://adbuymcp.com/how-to/install-the-mcp-server MCP access today is a private, single-tenant, self-hosted deployment behind a bearer token or your own tunnel. There is no public connector to add and no multi-tenant sign-in: establishing OAuth 2.1 for public multi-tenant MCP is an open launch gate, and the package deliberately does not embed an authorisation server. What follows sets up a deployment you control, exposing 76 tools over the same REST routes the browser uses. Before you start: Node.js, and the API running — locally it needs no credentials at all; A decision about auth: no token for local development, a service token for a private single-tenant deployment; For remote access, HTTPS and a hostname you can add to the allowlist Steps: 1. Run the API — Start the API in sandbox mode. With authentication disabled it needs no credentials and no environment variables, and the whole loop works. Watch out: That mode is explicitly not an authentication boundary — it grants its single dev identity every human permission implicitly. It is for local development and nothing else. 2. Build the MCP server — Build the MCP package so it emits its distributable, then point it at the API. The server talks to the API over HTTP only, with no database coupling. Watch out: It can point at any running API instance, which is what makes a separate staging deployment straightforward. 3. Register it with Claude Code over stdio — Add the server by pointing the client at the built stdio entry point. This is the recommended local setup. Watch out: Use the HTTP transport instead when the server runs somewhere other than the machine driving it. 4. Or register it over HTTP — Start the HTTP transport and register the URL with your client. The transport is stateless, so it works behind tunnels and load balancers with no session affinity. Watch out: Never expose the unauthenticated server. Set a bearer secret and pass it as an Authorization header. 5. Set the boundary before accepting traffic — Production requires exact allowed hosts and exact allowed browser origins — set origins to none for a deployment with no browser clients. Local execution binds to loopback unless a container deployment opts out explicitly. Watch out: Verifier time, request concurrency and JSON body size are all bounded, and a modern request missing its protocol version header is rejected before dispatch. 6. Check the health probe — The health endpoint returns service status and the API URL, with no authentication required. Watch out: If it answers but tool calls fail, the problem is almost always the API credential rather than the MCP server. Result: A working MCP server exposing 76 tools to your client, over the same routes, tenant boundary and approval gates the dashboard uses. ### Grant and revoke an agent mandate — https://adbuymcp.com/how-to/grant-and-revoke-an-agent-mandate Authority for an agent is a data structure rather than a setting. A mandate is a revocable, versioned grant enumerating exactly which actions are allowed, in which currency, for how long, and up to what money ceiling. The stronger guarantee sits underneath it: eleven consequential operations have no MCP tool at all — approval resolution, mandate replacement and revocation, mission closure, supplier dossier verification and live-grant issuance among them — so they cannot be reached by an agent regardless of what a mandate says. Before you start: A workspace and the human permissions that gate consequential actions, which are issued from server-controlled roles; A mission for the agent to work inside Steps: 1. Create a mission — A mission is a durable objective owned by your workspace. Everything the agent does happens inside one. Watch out: Agents cannot close their own missions. That is a human-only operation with no agent tool. 2. Grant a mandate with an explicit ceiling — Enumerate the allowed actions, the currency, the validity window and the money ceiling. A mandate that does not name an action does not permit it. Watch out: Set the ceiling to what you are willing to lose rather than what you intend to spend. It is the backstop, not the budget. 3. Let the agent prepare and propose — The agent works through ordered steps against the mandate, recording decisions and evidence as it goes. It can extract, compile, plan, draft creative and surface exceptions. Watch out: It cannot enable spend. Approval records evidence; enabling spend on a staged live line is a separately granted action. 4. Approve by exact digest — High-consequence work requires a single-use approval bound to an exact action digest. If the request that arrives does not match the bytes a human approved, it fails closed. Watch out: An integration test proves an agent attempting to resolve its own approval receives a 403 requiring human approval. This is enforced at the API rather than advertised in tool metadata. 5. Resolve exceptions yourself — An ambiguous supplier outcome raises an exception, halts the run, and waits for an explicit human resolution and resume. It is never retried automatically. Watch out: This is the failure this design exists to prevent: an agent that cannot read a supplier's response and tries again has just created a second campaign nobody knows about. 6. Revoke when the work is done — Revoke the mandate rather than leaving it to expire. Revocation is a human-only signed action with no agent tool. Watch out: A secret-only credential rotation also advances the connection revision and revokes issued grants, which is worth knowing before you rotate a key mid-campaign. Result: An agent that can do useful work inside an explicit ceiling, and cannot reach any of the operations that commit money. ### Connect your analytics — https://adbuymcp.com/how-to/connect-your-analytics The responded verb reads outcomes from analytics you already own — Plausible or GA4 — through a token you supply, matched to the campaign on an immutable UTM key. Connecting takes minutes. What takes longer, and what determines whether the data can be labelled observed rather than merely present, is verification: a human-verified dossier bound to the site and revision, and a known goal and revenue total reconciled before the source counts. Before you start: Admin access to your analytics property; A stats API token, and for Plausible the site identifier and its timezone; A landing page URL that will not change, because it becomes the campaign's immutable click destination Steps: 1. Save the credentials — Enter the site identifier, the timezone and the stats API token on the connections screen. Credentials are encrypted at rest with AES-256-GCM and are never returned by an API read. Watch out: For Plausible the timezone must be exactly Europe/London for the current reporting contract. A mismatch produces silently wrong day boundaries. 2. Fix the landing page before approving the plan — The plan's landing page becomes the campaign's immutable HTTPS click destination and its analytics identity. It cannot be changed afterwards. Watch out: URLs containing embedded credentials are rejected outright rather than passed to a provider or a browser. 3. Let the campaign key be appended — One non-blank campaign UTM key is preserved if you supply it; otherwise the campaign identifier is appended before the fragment without rewriting existing query bytes. Watch out: Blank, duplicate or brand-reused keys block approval, because two campaigns sharing a key make both unmeasurable. 4. Verify the goal exists, independently — Check in your analytics that the goal display name you configured actually exists, before interpreting any number that comes back. Watch out: The stats API cannot enumerate configured goals. Its query echo proves only that your exact display-name filter ran, so a zero may mean no conversions or may mean no such goal. 5. Reconcile before trusting it — Reconcile a known period's unique visitors, an exact configured goal display name, and revenue through the campaign selector's immutable key. Watch out: Until that reconciliation is done and bound to the site revision, the source is connected but not verified, and the distinction is the whole point of the provenance labelling. Result: Brand response measured on data you own and can audit, labelled observed rather than attributed. ### Approve creative for UK clearance — https://adbuymcp.com/how-to/approve-creative-for-uk-clearance UK broadcast and cinema advertising must be cleared before it airs, by separate bodies with separate rules. The three provisions that do most of the work: objective and superlative claims need documentary substantiation, medicines and health claims require evidence and are frequently blocked outright, and cinema copy containing strong language restricts pairing to 15 and 18 certificate films — a media constraint as much as a creative one. AdBuyMCP runs a deterministic pre-check that predicts these flags before you commission anything. Before you start: Draft copy or a script; Knowledge of what substantiation you actually hold for any claim in it; For cinema, an idea of which certificate films you want to sit alongside Steps: 1. Run the pre-check before production — The pre-check runs against the UK codes and returns findings with the rule named beside each one, rather than a bare pass or fail. It runs deterministically and needs no API key. Watch out: Running it after production is the expensive order. A flagged claim in a script costs an edit; the same claim in a finished film costs a reshoot. 2. Deal with superlatives first — Best, number one, market-leading and similar claims need documentary evidence. The pre-check finds the claim and names the rule. Watch out: Where you declare that you hold substantiation, the system records the declaration alongside the claim and states explicitly that it has not assessed the evidence. The clearance body will want the documents themselves. 3. Treat health claims as a stop — Medicines, medical devices, treatments and health claims require evidence and an unsubstantiated efficacy claim is routinely rejected. Watch out: This is the category where a rewrite is usually cheaper than an argument. 4. Check the cinema age rating implication — Strong language restricts your copy to 15 and 18 certificate films, which changes the audience you can reach and therefore the media plan. Watch out: Worth catching at the scripting stage, because it can invalidate the venue and film package the plan was built on. 5. Produce to the delivery spec — Broadcast needs a 16:9 ProRes 422 HQ mezzanine at 1920×1080 and 25fps, at least 20Mbps, supers inside title-safe and audio to EBU R128. Cinema needs a 2K flat DCP at 24fps with 5.1 audio, sixty seconds maximum. Watch out: A social cut will not clear on either. These are different masters, not different exports. 6. Submit and wait — Cinema routes through the distributor with a clearance certificate before print traffic. Broadcast clears before air. Watch out: Build the lead time into the flight. Around eighteen days is realistic for cinema once clearance is in the path. Result: A script whose predictable clearance problems were dealt with before anything was commissioned, and a media plan whose film and venue package is still valid after clearance. Not a guarantee of clearance: Clearcast, the CAA and the publishers decide, and the pre-check cannot assess substantiation. ### Set a budget across seven channels — https://adbuymcp.com/how-to/set-a-budget-across-seven-channels The single most common mistake in a multi-channel plan is spreading the budget so thin that no line can pace, and the floors exist to prevent it: search and DOOH at £100, CTV at £140 a week, audio and social at £250, cinema at £500 a week, addressable TV at £1,000. A line that cannot clear its floor is dropped from the plan and its budget redistributed across the lines that can. That is not a limitation to work around — it is the planner refusing to build something that cannot work. Before you start: A total monthly media budget; An objective, because it changes which channels are worth funding at all; A compiled persona, so the fidelity scores can inform the split Steps: 1. Check the budget clears the practical floor first — Below about £10,000 a month in total media this is the wrong product. A 10% fee on £2,000 is £200 against the buyer needing the most support from us, and a multi-channel plan at that level is too thin on every line. Watch out: This is the step where the honest answer is sometimes to buy search and social directly and come back later. 2. Read the fidelity scores before splitting — Fund the channels that can express your audience well. A 40 on cinema and an 85 on CTV are not equally good uses of the same pound for an audience-precision objective. Watch out: For a launch this reverses: cinema's attention profile may justify funding a channel with a low fidelity score, because reach and attention matter more than precision. 3. Concentrate rather than spread — Pick two or three channels and fund them to a weight that can pace and be measured, rather than putting a token line on all seven. Watch out: A seven-channel plan at £15,000 a month is usually worse than a three-channel one at the same budget, because none of the seven lines accumulates enough delivery to prove anything. 4. Let under-floor lines drop — The planner drops lines below their floor and redistributes, with an explicit note. Read the note rather than trying to force the line back in. Watch out: Vendor-side minimums are separate and bite later. Sky AdSmart's roughly £3,000 entry campaign is the one most likely to surprise a plan that cleared the £1,000 planner floor. 5. Account for the costs that are not media — Data segment uplift of roughly £2.50 to £3.00 per thousand on targeted business audiences, creative production, and the platform fee. Watch out: All of it appears on the plan before approval rather than in the reconciliation afterwards, which is the point of looking at it now. 6. Check whether the split can support the claim you want — If you intend to prove incrementality, the geography and spend have to support a powered test. That is worth establishing before funding rather than after. Watch out: A plan that cannot power a causal test can still produce reach and response evidence, and saying so in advance is a much better conversation than discovering it at reporting. Result: A plan whose funded lines can each pace, be measured, and support the claim you intend to make. ## 13a. Glossary — https://adbuymcp.com/glossary 76 terms, in 6 groups. Most glossaries in this industry restate a definition you could have found anywhere, which is why nobody cites them. This one does two things instead. Each entry opens with the neutral answer, written to be lifted whole and still be correct whoever is asking. Then it says what the term means for a UK buyer, and where we diverge from the norm — the take-rate entry names our ten per cent against the twenty per cent observed at The Trade Desk, the incrementality entry names the tests we refuse to run, and the entries on EUID and Acorn name capabilities we do not have. A reference page on a vendor's domain is not neutral, so the useful thing is to be specific rather than to pretend otherwise. Every term below is given twice: the neutral definition, which is true whoever is asking and is written to be quoted whole, and then what it means in practice for a UK buyer including where this product diverges from the category default. Where the two disagree with a claim made elsewhere in the industry, the divergence is the point. ### Channels and formats The seven things you can actually buy, and the words the sellers use for them. **Connected TV (CTV)** (also: connected television, streaming TV advertising, OTT advertising) — https://adbuymcp.com/glossary/ctv Advertising delivered inside streaming television on an internet-connected device: a smart TV, a streaming stick, a games console. It is bought like digital media, with audience targeting and impression-level reporting, but watched like television, full-screen and usually unskippable. In practice: CTV is the one channel in a UK plan where a business audience survives the translation almost intact, because the buying vocabulary can carry consent-chained B2B segments — Bombora intent topics, Dun & Bradstreet decision-maker segments — rather than only age and postcode. AdBuyMCP grades a compiled CTV plan at 85 out of 100 and person-level when at least one of those segments matches strongly, dropping to 65 and then 60 as the evidence weakens. That is the highest score anywhere in the system and it is still a projection: 85 means the best available proxy for a named audience, not a match on one. The segments carry a data cost of roughly £2.50 to £3.00 per thousand impressions, which appears on the plan rather than being absorbed into the price. **Addressable TV** (also: addressable television, Sky AdSmart, household addressable) — https://adbuymcp.com/glossary/addressable-tv Television advertising where different households watching the same programme at the same time receive different adverts, swapped in at the set-top box or in the stream. It is bought on household attributes and geography rather than on programme and time slot. In practice: In the UK this mostly means Sky AdSmart, with ITV and Channel 4 reachable through their own routes. Households are selected on Experian Mosaic groups and CACI attribute combinations intersected with postcode districts, which is why the honest fidelity grade is household rather than person: you are buying homes whose profile over-indexes on the attribute you asked for. AdBuyMCP grades it 55 out of 100, fixed — no amount of audience detail moves it, because the vocabulary cannot express more than that. Only an unmappable city moves it, and only downward, by fifteen points. It was the fastest-growing channel in this set at £1.84bn and 37% growth in 2025 on AA/WARC figures. We do not buy UK broadcast inventory today; these are booking-paperwork rails. **Linear TV** (also: broadcast TV, spot TV) — https://adbuymcp.com/glossary/linear-tv Traditional broadcast television bought as spots against programmes and time slots, where everyone watching that break sees the same advert. Priced on predicted audience delivery rather than on individual impressions. In practice: The distinction from addressable matters commercially rather than technically. A spot buys everyone watching; addressable buys the households you selected. That makes linear a reach instrument and addressable a targeting one, and it makes the fidelity question different — on linear there is no audience approximation to score, only an audience forecast to trust. UK linear inventory sits behind sales houses and is transacted as booking paperwork rather than through a self-serve API, which is why any platform promising one-click linear television in the UK is describing something else. **Digital out-of-home (DOOH)** (also: digital OOH, programmatic DOOH, digital billboards) — https://adbuymcp.com/glossary/dooh Advertising on digital screens in public places — roadside billboards, rail stations, shopping centres, office lifts, gyms. Increasingly bought programmatically by screen, venue type, geography and time of day rather than as a fixed two-week poster site list. In practice: DOOH is where the gap between honest and dishonest targeting language is widest. A screen cannot know who is in front of it; it can only be somewhere a cohort is known to be. AdBuyMCP compiles it to venue types, postcode polygons and dayparts and grades it 50 out of 100, geo-cohort — a statistical property of a place rather than of a person. Two things we will not claim: there is no movement-indexed screen ranking in it, because the compiler returns an empty ranking list, and no DOOH rail can be driven end to end from a compiled audience in live mode. UK DOOH was around £955m in 2025 on IAB and AA/WARC figures. **Out-of-home (OOH)** (also: outdoor advertising, billboards) — https://adbuymcp.com/glossary/out-of-home All advertising encountered outside the home: billboards, transport, street furniture and retail environments, printed or digital. Covers both classic printed sites and the digital screens that are steadily replacing them. In practice: Worth separating from DOOH when reading a media plan, because the two behave differently. Classic OOH is bought for a fixed period at a fixed price and cannot be paced, targeted by daypart or turned off. Digital screens can be. UK out-of-home overall was still contracting in the IPA Bellwether for the second quarter of 2026 even as the digital share of it grew, so an OOH line in a plan is usually a DOOH line in practice. **Cinema advertising** (also: cinema ads, on-screen advertising) — https://adbuymcp.com/glossary/cinema-advertising Advertising shown on cinema screens before a film, sold by audience packs, film genre and venue rather than by individual viewer. Delivered as a Digital Cinema Package and subject to its own age-rating clearance separate from broadcast clearance. In practice: Cinema carries the lowest fidelity score on this site — 40 out of 100, venue-level — and that number is published in the same size type as the 85 on CTV, because a site that only shows you its best number has told you nothing about its method. A cinema buy is a projection from catchment demographics onto a screen, and no UK seller can give you a person-level exposure record. It has two compensating properties: a full screen with the lights down has an attention profile nothing else in the plan matches, and the targeting involves no personal data at all, so it sits outside UK GDPR scope. It is booked as paperwork with roughly an eighteen-day lead time, not bought through an API. **Digital audio advertising** (also: streaming audio ads, online radio advertising) — https://adbuymcp.com/glossary/digital-audio Audio advertising delivered through streaming music, digital radio and podcasts, bought by show, genre, publisher segment and daypart. Distinct from broadcast radio in that individual impressions are counted and targeting is applied per listener session. In practice: Digital audio was around £410m in the UK in 2025 under the IAB's digital definition, growing about 14%. AdBuyMCP compiles it to show and genre allowlists plus publisher segments and grades it 65 out of 100, content-level — you are reaching people who listen to business podcasts in London, not the specific decision makers on your account list. One detail worth knowing: dayparts are set from the brief itself. A brief carrying firmographics buys the morning and evening commutes; one without them buys daytime and evening. That is a rule in the compiler rather than a model's guess, so you can disagree with it. **Podcast advertising** (also: podcast ads, host-read ads) — https://adbuymcp.com/glossary/podcast-advertising Advertising inside podcast episodes, either dynamically inserted at play time and targeted by listener, or read by the host and baked into the episode permanently. The two are bought, priced and measured differently. In practice: Dynamically inserted advertising is what a programmatic plan buys: targetable, reportable, and removable. Host-read is an endorsement, costs several times more, and cannot be turned off once published. UK self-serve minimums are low — Spotify's Ads Manager starts around £250 a campaign and Acast's self-serve at about $250, on desk research dated 3 August 2026 — while a host-read spot is closer to $5,000. Named-show buying happens through private marketplace deals, which are negotiated rather than clicked. **Paid search** (also: PPC, search advertising, SEM) — https://adbuymcp.com/glossary/paid-search Advertising against what someone typed into a search engine, bought by keyword and match type and usually paid per click. The only major channel where the buyer is responding to stated intent rather than inferring it. In practice: Search took £17.9bn of a £46.7bn UK market in 2025 on AA/WARC figures, and if you are reading this you are almost certainly already buying it competently. AdBuyMCP's contribution is not better search management — it is stopping search being planned in a different room from the television, audio and out-of-home around it, and deduplicating the audience across them. It is also the channel where the fidelity score moves most with the quality of your brief: buyer-authored intent topics score 82, interests without intent 68, and a persona label with no product context 45, with the plan telling you which you supplied. Nothing auto-buys: provider keyword suggestions have to be explicitly applied to a version-fenced draft. **Paid social** (also: social advertising, social media ads) — https://adbuymcp.com/glossary/paid-social Advertising inside social platforms, bought against platform-owned interest, demographic and behavioural audiences. The platforms optimise delivery inside their own inventory using signals no third party can see. In practice: Social was £11.5bn of the UK market in 2025 on AA/WARC figures. For most buyers, on its own, the platforms' own managers are free and better, and this site says so. The reason paid social is in a seven-channel product at all is deduplication and incrementality: without the channel carrying most of your budget, a cross-channel frequency number is wrong and an incremental-reach claim is unsupportable. AdBuyMCP holds one audience definition and lets Meta and TikTok each resolve it into whatever they actually offer, which avoids maintaining two divergent definitions — but exact availability is resolved per platform and the plan says so. Interest terms score 62, demographics alone 48, a bare city buy 38. **Programmatic advertising** (also: programmatic buying, automated media buying) — https://adbuymcp.com/glossary/programmatic Buying advertising through automated systems that decide, impression by impression, whether to bid and how much to pay for it. It describes the transaction mechanism rather than a channel: CTV, audio, DOOH and display can all be bought programmatically. In practice: The word does more damage than almost any other in this vocabulary, because it is used to imply that everything worth buying can be bought this way. In the UK it cannot. Cinema and addressable television transact as booking briefs, clearance steps and insertion orders — paperwork with lead times — and a platform describing those as programmatic is either wrong or reselling somebody else's paperwork. Of the twenty-five rails AdBuyMCP registers, six are booking workflows rather than APIs, and the supply page grades each one rather than blurring them together. ### Buying and commercials How media is transacted and where the money goes. The terms in this group are the ones a fee conversation turns on. **Demand-side platform (DSP)** (also: demand side platform) — https://adbuymcp.com/glossary/dsp Software an advertiser uses to buy advertising across many publishers through one interface, bidding on impressions in real time. It sits on the buyer's side of the market and charges a fee, usually a percentage of the media spent through it. In practice: Observed platform fees vary widely and are worth knowing before a negotiation: The Trade Desk's take rate was about 20% in 2024, DV360 sits around 10 to 15%, and the generic DSP tier runs 5 to 15%, on desk research dated 3 August 2026. We are not a DSP and own no inventory. It is the layer above them: it translates one audience into what each rail accepts, allocates budget between sellers that compete with each other, and reconciles what came back. It buys through StackAdapt, Amazon DSP and others as rails. If your channel set is one a DSP already covers well, a DSP is the simpler answer and this site says so on its comparison pages. **Supply-side platform (SSP)** (also: supply side platform) — https://adbuymcp.com/glossary/ssp Software a publisher uses to sell its advertising inventory to many buyers at once, running the auction and setting price floors. It sits on the seller's side of the same market a DSP buys from. In practice: The relevance to a buyer is the fee stack. A supply-side platform takes a fee and an exchange layer can take another on top of it, which is how the cumulative 'ad tech tax' arises without any single layer looking expensive. Knowing the SSP layer exists is what lets you ask the only question that matters about a media price: how much of this pound reaches the publisher? **Ad exchange** — https://adbuymcp.com/glossary/ad-exchange A marketplace in which advertising impressions are auctioned between demand-side platforms buying and supply-side platforms selling. It matches bids to inventory in the milliseconds between a page or stream starting to load and the advert appearing. In practice: Mostly invisible to a buyer, and mostly should be — except that each hop in the chain takes a margin, and the number of hops is rarely disclosed. Supply-path transparency, meaning a published account of which intermediaries a bid passes through, is something the largest advertisers now demand and smaller ones almost never get. **Real-time bidding (RTB)** (also: RTB) — https://adbuymcp.com/glossary/real-time-bidding The auction mechanism underneath most programmatic advertising: when an impression becomes available, buyers are invited to bid on it and the winner's advert is served, all within roughly a hundred milliseconds. In practice: Two consequences a buyer feels. Price is set per impression rather than per campaign, so a plan's CPM is a forecast rather than a rate. And the auction is where most of the ad tech fee stack is levied, which is why the cumulative take between advertiser and publisher can be far larger than any single platform's disclosed rate. **Private marketplace (PMP)** (also: PMP, private auction, deal) — https://adbuymcp.com/glossary/private-marketplace A negotiated arrangement in which a publisher makes specific inventory available to specific buyers at agreed terms, transacted programmatically through a deal identifier rather than in the open auction. In practice: This is how named-show podcast buying and most premium audio actually happens: you do not target a show, you negotiate a deal and then execute it programmatically. AdBuyMCP's audio compiler emits a named-show package as a deal specification rather than as a targeting parameter, which is the honest shape — the deal has to exist before the plan can spend against it. **Deal ID** — https://adbuymcp.com/glossary/deal-id The identifier that carries a negotiated private-marketplace arrangement into a programmatic buying system at auction time, telling the auction which buyer gets access to which inventory at which agreed price. In practice: Worth knowing because it is the practical boundary between what a platform can buy on your behalf today and what requires a conversation first. Our compiled plans carry an empty deal list: deals arrive when a buyer hand-picks inventory or when one has been negotiated, not from a compiled audience. **CPM** (also: cost per mille, cost per thousand) — https://adbuymcp.com/glossary/cpm The cost of one thousand advertising impressions, and the standard price unit for advertising bought on attention rather than on action. Television, streaming, audio, out-of-home and cinema are all quoted this way. In practice: Two cautions. A CPM in a plan is a forecast, not a rate, because most inventory is auctioned per impression. And a blended CPM computed by averaging two days of averages is a fabricated number — AdBuyMCP's delivery payload refuses to produce one, keeping ratios and cost averages as dated observations that are never summed. Data segments add their own CPM uplift on top of the media: roughly £2.50 per thousand for Bombora topics and £3.00 for Dun & Bradstreet segments, shown on the plan rather than absorbed. **CPC** (also: cost per click) — https://adbuymcp.com/glossary/cpc The cost of one click on an advert, and the standard price unit for paid search and for much of paid social, where the advertiser pays for a response rather than for the impression that produced it. In practice: A CPC transfers delivery risk to the seller, which is why it dominates the channels where clicks are plentiful and is almost unknown on television, cinema and out-of-home. Comparing a CPC channel to a CPM channel on cost per click is a category error: it flatters search and tells you nothing about whether the other channels created the demand search then harvested. **CPA** (also: cost per acquisition, cost per action) — https://adbuymcp.com/glossary/cpa The cost of one completed action, whether that is a sale, a lead or a booking, calculated by dividing spend by the actions attributed to it. That makes it entirely dependent on the attribution model underneath. In practice: The most quoted and least comparable number in media. Two CPAs are only comparable if they share an attribution window, a model and a definition of the action, and they almost never do. A CPA computed on last-click will always flatter search and starve everything that created the demand. This is the number an incrementality test exists to check. **Cost per outcome (CPO)** (also: guaranteed outcomes, outcome-based pricing) — https://adbuymcp.com/glossary/cost-per-outcome A pricing model in which the advertiser pays only when a defined outcome occurs — an install, a purchase, a store visit — with the platform carrying the delivery risk and embedding its margin in the outcome price. In practice: The most aggressive pricing innovation in this category. tvScientific offers it on connected TV, restricted to qualifying advertisers. It is a genuinely better offer than a percentage of media for a performance buyer who qualifies, and this site says so on the comparison page rather than pretending otherwise. It requires measurement maturity and risk capital, and we do not offer it. **Take rate** (also: platform fee, tech fee) — https://adbuymcp.com/glossary/take-rate The share of an advertiser's media budget retained by the platform rather than spent on media, whether that share is disclosed or embedded in the price. The word is neutral about disclosure, which is why the useful question is the figure rather than the policy. In practice: The single most useful number to ask any media platform for, and the one most often deflected. Observed rates on desk research dated 3 August 2026: The Trade Desk about 20%, DV360 around 10 to 15%, generic DSPs 5 to 15%, UK agencies 10 to 30% with minimum monthly fees of £500 to £1,000. MNTN's own documentation states it keeps an undisclosed percentage inside the CPM. AdBuyMCP charges 10% of media, falling to 5% plus £749 a month above £25,000 of trailing thirty-day spend, with one ledger entry accruing per delivery row so the fee sits beside the media it was charged on. That is a contractual commitment and a design property; it is not yet a demonstrated practice, because no rail has ever spent. **Working media** — https://adbuymcp.com/glossary/working-media The portion of an advertising budget that actually buys advertising, as opposed to the portion consumed by platform fees, data costs, agency fees, ad serving and production. Usually expressed as a percentage of total spend. In practice: The number that makes fee comparisons real. A 10% platform fee on top of a 15% data cost and an agency retainer is not a 10% cost, and every intermediary between the advertiser and the publisher takes its share before the money arrives. This site publishes no figure for how much of a typical pound survives that journey, because it has no sourced one. Any platform that cannot tell you its own working-media percentage is telling you something. **Ad tech tax** (also: supply chain fees) — https://adbuymcp.com/glossary/ad-tech-tax The cumulative share of an advertiser's budget taken by the intermediaries between advertiser and publisher — demand platform, exchange, supply platform, data vendors, verification — none of which individually looks large. In practice: It is cumulative and mostly undisclosed, which is why it is the category's most persistent grievance. The defence is not a lower headline rate; it is supply-path transparency and a fee that appears on the line of media it was charged against rather than in a quarterly summary. **Arbitrage** (also: spread, media markup, principal trading) — https://adbuymcp.com/glossary/arbitrage Buying media at one price and reselling it to the advertiser at a higher one, keeping the difference undisclosed. Distinct from a disclosed fee in that the advertiser cannot see what the media actually cost. In practice: Legal, common, and the reason 'transparent' is a loaded word in this industry. The test is simple: can you see what the media cost, separately from what you were charged? AdBuyMCP's fee engine is built with no spread — one ledger entry per delivery row carrying that row's date, campaign, channel, spend and fee — and the terms commit to media passed through at cost with no undisclosed margin or rebate. Being precise about the status of that commitment: it is contractual and architectural, not yet demonstrated, because no rail has ever spent and therefore no invoice has ever been reconciled against it. **Insertion order (IO)** (also: IO, booking) — https://adbuymcp.com/glossary/insertion-order A contractual document committing an advertiser to buy specified advertising from a specified seller at agreed terms. The mechanism by which media that is not bought programmatically is bought instead. In practice: The word that separates a channel you can buy in an afternoon from one that takes a fortnight. UK cinema and addressable television both transact this way: a brief goes to the sales house, clearance happens, an order is confirmed, and delivery is reconciled afterwards. AdBuyMCP automates the paperwork — generating the brief with the right audience packs, driving the clearance step, holding the state machine — but it does not make the booking and it does not report the delivery. Every live step is acknowledged by a named person quoting the sales house's own reference, and no live step advances on a timer. **Rate card** — https://adbuymcp.com/glossary/rate-card A seller's published list prices for its advertising inventory, before any negotiation, discount or volume agreement is applied to them. In practice most large media is bought below rate card, which makes the card a starting position rather than a price. In practice: Relevant to a small advertiser mainly as a warning: the discount available depends on the volume behind them, which is one of the genuine arguments for buying through an intermediary rather than direct. It is also the reason a published platform fee and a published rate card are different kinds of promise. ### Targeting and identity How an audience is described to a seller, and how much of the description survives. Most of the honest difficulty in media buying lives here. **Audience segment** (also: segment, audience) — https://adbuymcp.com/glossary/audience-segment A named group of people a seller will let you target, defined by some combination of demographics, behaviour, stated interest or inferred intent, and identified in a buying system by a specific identifier. In practice: The critical thing about a segment is that it is somebody's model of a group, not the group. 'IT decision makers' is a set of people a data vendor believes are IT decision makers, built from signals of varying quality. AdBuyMCP's design decision here is structural: the language model extracts semantics from your brief and never emits a vendor segment identifier. Deterministic registry tables do that, matching your terms against catalogue keywords by whole-token overlap. That is why a compiled plan can be audited line by line, and why a hallucinated segment cannot reach a buy. **First-party data** — https://adbuymcp.com/glossary/first-party-data Data an organisation has collected directly from its own customers and visitors — a CRM list, purchase history, site behaviour. Distinguished from third-party data by the existence of a direct relationship with the person it describes. In practice: The most valuable targeting asset most advertisers own and the hardest to activate on premium channels, because getting a list from your database onto a television requires an identity bridge that does not compromise the list. Consent-chained identifiers such as EUID exist for exactly this. Being accurate about the current state: AdBuyMCP's compiled CTV plans carry an empty EUID audience list, so that bridge is designed rather than in use, and the adapter behind it is a fail-closed stub pending an operator agreement. **Third-party data** — https://adbuymcp.com/glossary/third-party-data Audience data collected by one organisation and licensed to another for targeting, with no direct relationship between the licensee and the people described. Sold as segments through data marketplaces attached to buying platforms. In practice: It carries a real cost and a real provenance question. The cost appears as a CPM uplift on top of the media — around £2.50 per thousand on Bombora topics, £3.00 on Dun & Bradstreet. The provenance question is what a lawful-basis manifest exists to answer: which provider, on what basis, with which conditions still owed by you. Consent-chained means the provider asserts a consent chain, not that anyone downstream has verified every link in it, and this site does not claim otherwise. **Contextual targeting** — https://adbuymcp.com/glossary/contextual-targeting Targeting based on the content an advert appears beside rather than on data about the person seeing it — a show, a genre, a page subject, a venue type. Requires no personal data about the audience. In practice: Structurally privacy-safe and increasingly the honest default as behavioural identifiers decline. It is what AdBuyMCP's audio compilation actually is: show and genre allowlists proxying the audience, graded content-level at 65 rather than dressed up as person-level. The same logic makes cinema and DOOH sit outside UK GDPR scope for delivery — a venue catchment is a property of a place, not of a person. **Lookalike audience** (also: similar audience, act-alike) — https://adbuymcp.com/glossary/lookalike-audience An audience built by finding people who resemble a seed group — usually the advertiser's existing customers — on whatever signals the platform holds. The resemblance model is the platform's and is not disclosed. In practice: Effective and opaque in equal measure, and available only where a platform has both your seed list and enough signal to model against it, which in practice means the large social platforms. It is worth remembering that a lookalike inherits every bias in the seed: modelling on your existing customers reliably finds more people like your existing customers, which is not the same as finding growth. **Retargeting** (also: remarketing) — https://adbuymcp.com/glossary/retargeting Showing advertising to people who have already interacted with a brand — visited a site, abandoned a basket, opened an email. Depends on being able to recognise the same person again on a later occasion. In practice: The channel where attribution flatters itself most, because retargeting reaches people who were already going to convert. It usually looks like the best-performing line in a plan and is usually the one most in need of an incrementality test. A geo-lift or holdout design is the only reliable way to find out how much of it was incremental. **Frequency capping** (also: frequency cap) — https://adbuymcp.com/glossary/frequency-capping Limiting how many times one person or household sees the same advert within a given period, to control wear-out and wasted impressions. It is set per campaign, per channel or — rarely — across channels, and that granularity is what decides whether the cap means anything. In practice: Almost every platform caps within itself and almost none caps across sellers, which is why a multi-channel plan quietly overexposes the same households. Capping across channels requires one exposure record keyed on something all the channels can supply. AdBuyMCP's exposure ledger is built for exactly that, on four keys of decreasing precision. Being exact about its state: the spine is designed, schema-backed and unit-tested, and is currently fed only by the deterministic sandbox — a live report proves aggregate delivery, not household exposure, so the system refuses to mint identities from it. **Reach** — https://adbuymcp.com/glossary/reach The number of distinct people or households exposed to a campaign at least once, as opposed to the total number of impressions delivered. Usually expressed as a count or as a percentage of a defined universe. In practice: Reach across channels is almost always modelled rather than counted, because no single seller can see the others. AdBuyMCP models it with Sainsbury deduplication against the persona universe and labels the result modelled wherever it appears — the label is a permanent property of the field rather than a footnote. The naive alternative, adding each channel's reach together, is the single most common inflated number in a media report. **Deduplication** (also: dedup, cross-channel deduplication) — https://adbuymcp.com/glossary/deduplication Removing the double-counting that occurs when the same person is reached by more than one channel or placement in a campaign, so that combined reach reflects distinct people rather than the sum of each channel's separate count. In practice: The reason a cross-channel plan needs one exposure record rather than seven reports. Without it, a launch that buys the same households on television, streaming and out-of-home reports three audiences that are substantially one audience, and the number that reaches a board is wrong by a wide and unknowable margin. Deduplication at launch matters more than at any other moment, because a launch buys the same people repeatedly by design. **Share of voice** (also: SOV) — https://adbuymcp.com/glossary/share-of-voice A brand's advertising presence expressed as a proportion of all advertising in its category, in a given market and period. Used as a planning benchmark on the argument that share of voice above share of market tends to precede growth. In practice: Useful for setting a budget's order of magnitude and useless for judging a campaign, because it is a share of spend rather than a share of attention. It is worth knowing mainly because it is the reasoning behind most large brand budgets, and therefore the frame a buyer arrives with when they ask what a channel costs. **Experian Mosaic** (also: Mosaic) — https://adbuymcp.com/glossary/experian-mosaic A UK geodemographic classification that assigns every household to one of a set of named types based on demography, affluence, life stage and location. Used to select households for addressable television and direct media. In practice: The vocabulary UK addressable television is actually bought in. A brief naming senior business decision makers compiles into Mosaic groups and occupation attributes intersected with postcode districts — which is why the honest grade is household rather than person, and why AdBuyMCP scores addressable TV 55 out of 100. You are buying homes whose profile over-indexes on the attribute you asked for. **CACI Acorn** (also: Acorn) — https://adbuymcp.com/glossary/caci-acorn A UK consumer classification that segments postcodes into named types by demographics, lifestyle and behaviour, used to rank areas rather than people. In practice that means catchment work: which cinema, which retail park, which high street. In practice: The intended basis for ranking cinema venues by how far their catchment over-indexes on a target audience. Stated plainly: we do not hold that licence. Cinema venue lists are currently built from the persona's own postcode districts, the catchment over-index figures shown on a sandbox booking brief are deterministic fixtures rather than data, and the Acorn adapter is a fail-closed stub. The supply page grades it accordingly. **Mobile advertising ID (MAID)** (also: MAID, IDFA, GAID, advertising identifier) — https://adbuymcp.com/glossary/maid A resettable identifier assigned by a mobile operating system to a device, used to recognise the same device across apps for advertising. Apple's IDFA and Google's advertising ID are the two that matter. In practice: A declining asset. Platform changes have reduced availability sharply and the direction is one way, which is why any measurement spine built on advertising identifiers alone degrades on a schedule somebody else controls. AdBuyMCP holds hashed advertising identifiers for at most ninety days, after which the hash is persistently stripped from the exposure event, re-grading it to a coarser key rather than deleting the record. That rule lives in the key-translation service, which is the only module we have that performs identity joins at all. **European Unified ID (EUID)** (also: EUID) — https://adbuymcp.com/glossary/euid A consent-based identifier for European advertising, derived from a hashed email or phone number provided with consent, designed to let first-party audiences be addressed across publishers without third-party cookies. In practice: The intended bridge from a customer list to premium inventory under UK and EU data protection. To be exact about its state here: AdBuyMCP's CTV compiler emits an empty EUID audience list, so no compiled plan carries one today, and the adapter is explicitly non-transacting pending an operator agreement. Anyone reading elsewhere that we target CTV on EUID should know that identifier reaches no plan yet. **Unified ID 2.0 (UID2)** (also: UID2) — https://adbuymcp.com/glossary/uid2 An open-source, consent-based advertising identifier derived from a hashed email or phone number, developed as an industry alternative to third-party cookies. EUID is its European counterpart. In practice: Mentioned here mainly so the family is legible: UID2 is the North American implementation and EUID the European one, and a vendor discussing one may mean either. Neither is in use here today. **Household graph** — https://adbuymcp.com/glossary/household-graph A dataset linking devices, identifiers and addresses believed to belong to the same household, allowing a television, a laptop and a phone in one home to be recognised as one buying unit. In practice: What makes deduplication between connected TV and addressable television possible at all, since neither can see an individual. It is also the point where a privacy review should press hardest, because a household graph is an inference about people who never supplied it. AdBuyMCP's ledger keys on an opaque household key issued by a separate key-translation service and never resolves it back to anything. **Identity resolution** — https://adbuymcp.com/glossary/identity-resolution The process of deciding that two different identifiers refer to the same person or household, so that behaviour seen under one can be joined to the other. In practice: The highest-risk operation in advertising data and the one worth isolating architecturally. In AdBuyMCP exactly one module performs identity joins — the key-translation service — and the exposure ledger every channel writes into never resolves a key back to a person or across key spaces. That boundary is why the ninety-day retention rule has one place to be enforced and one place to be audited. **TCF consent string** (also: Transparency and Consent Framework, TCF v2.3) — https://adbuymcp.com/glossary/tcf A standardised, encoded record of what a user consented to, produced at the point of collection and readable by every participant downstream. It is passed along the advertising supply chain so that each participant can check whether it has a basis to process that person's data before it does. In practice: The mechanism by which 'consent-chained' becomes checkable rather than asserted. AdBuyMCP's policy engine attaches a standing condition to every consent-chained data source requiring that the supply chain be framework-validated and that the vendor attestation be verified on connection. That is a condition you still owe rather than a box already ticked, which is the honest description of what a lawful-basis manifest produces. ### Measurement and evidence What you can prove afterwards, and the difference between a number that means something and a number that does not. **Incrementality** (also: incremental lift, true lift) — https://adbuymcp.com/glossary/incrementality The share of an outcome that happened because of the advertising, measured against what would have happened anyway. Measured by comparing an exposed group with a comparable unexposed one, not by counting conversions that followed an advert. In practice: Sold by most platforms in this category and refused by almost none, which is the difference worth caring about. AdBuyMCP runs a matched-market geo-lift test and puts a power analysis in front of it at 5% significance and 80% power. If the design cannot detect a plausible lift, it returns an underpowered verdict, the required sample days and the minimum detectable effect — and no lift figure at all. Its own words: reporting a number from it would be reporting noise. Nothing here has produced a causal result for a customer, because nothing has run live. **Geo-lift test** (also: matched market test, geo experiment) — https://adbuymcp.com/glossary/geo-lift An experiment that runs advertising in some geographic markets and withholds it from comparable others, then compares outcomes between them. The causal design that needs no login, pixel or match rate. In practice: The most practical causal method available to a brand advertiser, and the one that suits out-of-home, cinema and television — channels where person-level measurement is impossible. It needs enough independent markets, enough spend and enough conversion volume, which is why a twenty-catchment retailer can usually power one and a single national campaign usually cannot. AdBuyMCP also refuses a contaminated design: if every candidate control market received delivery from the campaign, none is a holdout, and it says so and stops. **Holdout** (also: control group) — https://adbuymcp.com/glossary/holdout A group deliberately excluded from seeing a campaign and kept as a comparison against the group that was exposed to it. Without one, the counterfactual — what would have happened anyway — has to be assumed rather than observed. In practice: The most common way a lift test quietly measures nothing is a contaminated holdout: the control markets received some of the campaign, so the comparison is exposed-versus-slightly-less-exposed. AdBuyMCP checks for this before the readout rather than after, and refuses with a specific instruction — exclude those markets from targeting for the test window, or nominate a control the campaign does not buy. **Matched market** — https://adbuymcp.com/glossary/matched-market A control market chosen because its historical outcome pattern closely tracks the test market's, so that a later divergence can be attributed to the campaign. In practice: Matching is done on the pre-campaign series rather than on demography, because two cities that look alike on paper often behave differently commercially. Getting this wrong is invisible in the output: a badly matched pair produces a confident number that means nothing, which is why the pairing and the power analysis both have to happen before the media runs. **Difference-in-differences** (also: diff-in-diff, DiD) — https://adbuymcp.com/glossary/difference-in-differences A method that compares the change in an exposed group against the change in a control group over the same period, so that anything moving both in step — seasonality, a competitor, the weather — cancels out. In practice: The readout underneath a geo-lift result. It is worth understanding because it explains what the test is actually robust to: shared shocks cancel, but anything that hit only the test markets does not. That is the residual risk in any geo experiment and the reason a result should carry a confidence interval rather than a single number. **Statistical power** — https://adbuymcp.com/glossary/statistical-power The probability that a test will detect a real effect of a given size, if an effect of that size genuinely exists in the population being measured. A design with low power will usually miss a genuine effect, raises the share of the results it does produce that are noise rather than signal, and exaggerates the size of the effects it does detect. In practice: The concept that separates measurement from theatre, and the one almost no media platform surfaces. AdBuyMCP computes it before running a lift test, defaulting to 80% power at 5% significance on daily observations, and refuses to proceed when the design falls short — returning the days of post-exposure data the test would actually need and the minimum lift it could detect. The refusal is the feature: an underpowered result is not a weak answer, it is a random one, and a random number in a board pack is indistinguishable from a real one. **Minimum detectable effect (MDE)** (also: MDE, minimum detectable lift) — https://adbuymcp.com/glossary/minimum-detectable-effect The smallest true effect a test could reliably detect, given its sample size, the variance in the data and the power the design was set to. Effects smaller than the MDE will usually be missed even when they are real. In practice: The number that tells you whether a test is worth running before you run it. If your design's MDE is a 30% lift and a realistic lift is 5%, the experiment cannot answer your question and should not be commissioned. AdBuyMCP returns the MDE alongside an underpowered verdict precisely so the answer is actionable — you learn what would have to change rather than only that it failed. **Attribution** — https://adbuymcp.com/glossary/attribution Assigning credit for an outcome to the advertising touchpoints that preceded it, according to a rule the advertiser chose. A description of correlation in a conversion path, not a demonstration that the advertising caused anything. In practice: The distinction between attribution and incrementality is the most consequential one in media measurement, and most reporting blurs it. AdBuyMCP keeps them in separate verbs: responded covers observed outcomes on your own property matched to the campaign's immutable key, and carries an explicit note that it is correlation; caused is reserved for a powered geo-lift result. Two exclusions are enforced rather than advised — the synthetic baseline series never inflates response counts, and nothing dated before the flight can count as a response to it. **Last-click attribution** — https://adbuymcp.com/glossary/last-click An attribution rule that gives all the credit for an outcome to the final advertising touchpoint before it, and none to the rest. Simple, universally available, and systematically biased towards whichever channel sits closest to the conversion. In practice: It reliably over-credits search and retargeting and under-credits everything that created the demand they harvested, which is why a business optimising purely on last-click tends to shrink towards its existing customers. It is not useless — it is a consistent operational signal — but it cannot answer whether a channel was incremental, and no amount of modelling on top of it turns it into an answer. **Marketing mix modelling (MMM)** (also: MMM, media mix modelling, econometrics) — https://adbuymcp.com/glossary/marketing-mix-modelling A statistical method that estimates each channel's contribution to an outcome by regressing historical sales against historical spend and external factors. Works without personal data and covers channels that cannot be tracked individually. In practice: Attractive because it needs no identifiers and covers television, cinema and out-of-home. Limited because it is correlational, hungry for history, and sensitive to specification — two competent modellers can produce materially different answers from the same data. AdBuyMCP labels anything modelled as modelled and keeps it strictly separate from a powered causal result; a modelled estimate presented as evidence of causation is exactly the failure its four-verb model exists to prevent. **Brand lift** (also: brand lift study, recall study) — https://adbuymcp.com/glossary/brand-lift A survey-based measurement comparing exposed and unexposed people on awareness, recall, consideration or favourability, to estimate what a campaign changed in attitude rather than in behaviour. In practice: The natural measure for a brand launch and the fourth verb in AdBuyMCP's measurement model — and it is not built. The API returns it with the status placeholder and a note that no study has been commissioned, and the panel vendor's adapter is a fail-closed stub pending a contract. It appears on this site anyway, at full weight, because a four-verb model that quietly ships three verbs is precisely the kind of rounding we exist not to do. **Viewability** — https://adbuymcp.com/glossary/viewability Whether an advert was actually rendered in a position where a person could have seen it, measured against an industry threshold — commonly half the pixels for one second for display, two seconds for video. In practice: A served impression and a viewable impression are different things, and the gap is where a meaningful share of digital budgets goes. Stated plainly: AdBuyMCP has no independent verification integration — no viewability, brand-safety or fraud vendor is wired in — and describes that as unbuilt rather than unnecessary. **Proof of play** — https://adbuymcp.com/glossary/proof-of-play A seller-supplied record confirming that a specific advert played on a specific screen at a specific time, with the play logged rather than estimated. The out-of-home equivalent of an impression log, since a screen cannot report who was in front of it. In practice: The strongest delivery evidence out-of-home can produce, and the reason a DOOH report proves the media ran rather than that anyone saw it. It is a delivery fact, not an audience fact, and a plan that treats proof-of-play counts as reach is overstating by whatever the seller's audience model says. **VAST** (also: Video Ad Serving Template) — https://adbuymcp.com/glossary/vast The industry XML standard that tells a video player where to fetch an advert, how long it runs and which tracking events to fire. The plumbing underneath most connected TV and online video advertising. In practice: Relevant to a buyer mainly as a delivery requirement: a CTV creative is not a video file, it is a video file plus its trafficking metadata, and a mismatch between them is a common reason a campaign fails to start. AdBuyMCP generates storyboards with the metadata attached so the creative and its trafficking record are one object rather than two things to reconcile. **Evidence provenance** (also: provenance label, evidence label) — https://adbuymcp.com/glossary/provenance A label attached to a number recording how it was produced — reported by the seller, observed on your own property, estimated by a model, or established by a powered experiment — so its strength travels with it. In practice: The mechanism that keeps four measurement verbs from collapsing into one flattering number. AdBuyMCP uses four labels: provider-reported, observed, modelled and powered causal, and only the last supports the word caused. The governing rule is blunt — no provenance label may advance on the strength of sandbox or API-shaped data — which is why no sandbox campaign can ever reach a causal grade, and none has. **Modelled** (also: modeled, estimated) — https://adbuymcp.com/glossary/modelled A number produced by a formula or an estimate rather than counted directly from observed events, and therefore carrying assumptions with it. Deduplicated reach and frequency across channels are the common examples, since no single seller can observe the others. In practice: The word matters because a modelled number and a counted one look identical in a report. AdBuyMCP's reach and frequency fields carry the label permanently rather than in a footnote — the field's type includes it — so a modelled estimate cannot be quoted as an observation by accident. ### Compliance and clearance UK data protection and UK advertising clearance: what has to be true before an ad runs. **UK GDPR** (also: General Data Protection Regulation, data protection) — https://adbuymcp.com/glossary/uk-gdpr The United Kingdom's data protection regime, retaining the structure of the EU General Data Protection Regulation. Requires a lawful basis for processing personal data and gives individuals rights over data about them. In practice: For a media buyer the practical questions are which targeting involves personal data at all, on what basis, and for how long it is kept. Two of AdBuyMCP's seven channels — cinema and DOOH — involve no personal data in delivery and are recorded as outside scope rather than left to be inferred. Where personal data is involved, the compiled plan carries a manifest naming each source and its basis. Note what that is and is not: an engineering artefact that tells you what you still owe, not a compliance guarantee. This site does not claim to be UK GDPR compliant on your behalf. **Lawful basis** — https://adbuymcp.com/glossary/lawful-basis The legal ground relied on to process personal data, of which six exist under UK GDPR and at least one must be identified before processing begins. In advertising the two that matter are consent and legitimate interests, and the basis has to be settled before the processing, not after. In practice: Different segments in the same media plan can rely on different bases, which is why recording it per segment rather than per campaign is the only useful granularity. AdBuyMCP's policy engine assigns a basis per data source and attaches the condition that comes with it — a framework-validated supply chain for consent-chained sources, a completed assessment on file for legitimate-interest ones. **Legitimate interests** — https://adbuymcp.com/glossary/legitimate-interest A lawful basis permitting processing necessary for interests pursued by the controller or a third party, provided those interests are not overridden by the rights of the individual. Requires a documented balancing assessment. In practice: The basis most advertising targeting actually relies on, and the one most often asserted without the assessment that makes it valid. AdBuyMCP attaches a standing condition to every legitimate-interest source stating that an assessment must be on file for it — which is a statement of what you owe rather than confirmation that it exists. **Consent** — https://adbuymcp.com/glossary/consent A freely given, specific, informed and unambiguous indication that someone agrees to their data being processed. Must be as easy to withdraw as to give, and cannot be inferred from silence or inactivity. In practice: In advertising, consent is usually collected far upstream by a publisher or a data vendor and passed down the chain as an encoded string. 'Consent-chained' means a provider asserts that chain exists — not that anyone downstream has verified every link in it, and this site is careful not to imply otherwise. Withdrawal propagation, meaning that a withdrawal upstream actually stops downstream processing, is the part most implementations handle worst. **Legitimate Interests Assessment (LIA)** (also: LIA) — https://adbuymcp.com/glossary/lia The documented three-part test of purpose, necessity and balancing that underpins any reliance on legitimate interests. It must be completed and retained before an organisation may rely on legitimate interests as its lawful basis for processing personal data. In practice: The document a data protection officer will ask for first when a plan relies on legitimate interests, and the one most advertisers discover they do not have. AdBuyMCP surfaces the requirement as a condition on the compiled plan rather than leaving it to be discovered in a review, but producing the assessment is the advertiser's obligation. **Data Protection Impact Assessment (DPIA)** (also: DPIA) — https://adbuymcp.com/glossary/dpia A structured assessment of the privacy risks of a processing activity, required where processing is likely to result in a high risk to individuals — which large-scale profiling for advertising generally is. In practice: Named here for completeness and for one honest disclosure: no completed impact assessment exists for AdBuyMCP's exposure ledger, and it is listed on our own checklist as outstanding rather than done. The machinery that would feed one — per-source lawful basis, per-key legal grading, a documented retention rule — exists; the assessment itself does not. **Clearcast** — https://adbuymcp.com/glossary/clearcast The body that pre-approves television advertising in the UK on behalf of the major broadcasters, checking copy against the broadcast advertising code before it can air. Approval is required, not advisory. In practice: The practical constraint is time and evidence. Superlative and market-leadership claims need documentary substantiation, and health claims are routinely rejected without it. AdBuyMCP's creative pre-check predicts which claims will be flagged before you commission the asset — and states clearly that it cannot read a substantiation document and does not assess evidence quality. Where you declare that you hold evidence, both the claim and the declaration stay on the record for the body that actually decides. **Cinema Advertising Association (CAA)** (also: CAA) — https://adbuymcp.com/glossary/caa The body that clears advertising for UK cinema, issuing a certificate and an age rating that determines which films an advert may be shown alongside. Separate from broadcast clearance and required before print traffic. In practice: The age rating is a media constraint as much as a creative one: copy containing strong language restricts pairing to 15 and 18 certificate films, which changes the audience you can reach. That is worth knowing at the scripting stage rather than after production, which is why the pre-check flags it there. **BCAP and CAP codes** (also: BCAP, CAP code, advertising code) — https://adbuymcp.com/glossary/bcap The two UK advertising codes, enforced by the Advertising Standards Authority: BCAP governs broadcast advertising and CAP governs everything non-broadcast. They set the rules on substantiation, harm, offence and specific product categories. In practice: Three provisions do most of the work in this category. Objective and superlative claims need documentary substantiation. Medicines, medical devices and health claims require evidence and are frequently blocked outright. And rules on specific categories — alcohol, gambling, financial products — constrain both copy and placement. AdBuyMCP's pre-check names the rule alongside each finding rather than returning a bare pass or fail. ### AdBuyMCP's own vocabulary Six terms we either invented or use more strictly than the category does. Defined here so nobody has to guess. **Fidelity score** — https://adbuymcp.com/glossary/fidelity-score A published score out of 100 answering one question: how much of an audience definition survives translation into a channel's native targeting words. It is a property of that translation rather than a forecast, and it is not a quality rating. In practice: AdBuyMCP's own term. Scores run from 85 on connected TV down to 40 on cinema, and both appear in the same size type, because a page that shows only its best number has explained nothing about its method. Grades run person, household, content, geo-cohort and venue, each one step further from the individual. Four of the seven channels carry a fixed constant — audio 65, addressable TV 55, DOOH 50, cinema 40 — because those vocabularies cannot express more however good the brief. Three vary across three discrete values: CTV 85, 65 or 60; search 82, 68 or 45; social 62, 48 or 38. Naming a city the UK geography table cannot map costs 15 points on every channel, and the rationale names the city. **Lawful-basis manifest** — https://adbuymcp.com/glossary/lawful-basis-manifest A record attached to every compiled channel plan naming each data source used, the lawful basis it relies on, the conditions that come with it, and the warnings raised while compiling — produced automatically rather than assembled on request. In practice: It exists so the answer to 'where did this audience come from?' is a document a data protection officer can read rather than a conversation, and so that the record a reviewer sees is the same record the compiler acted on. Its limits should be stated as plainly as its purpose: it attaches conditions you still owe, such as requiring an assessment on file or verifying a vendor attestation on connection. It tells you what remains to be done. It is not a compliance certificate and turning its conditions into enforced launch gates is work we have not finished. **Persona compiler** — https://adbuymcp.com/glossary/persona-compiler The deterministic stage that turns one described audience into each channel's native targeting specification — segment identifiers, attribute combinations, show allowlists, venue packs, keyword seeds — together with a fidelity score and a lawful-basis manifest. In practice: The division of labour is the whole design. A language model extracts semantics from the brief: what kind of person, which interests, which geography, what intent. It never emits a vendor segment identifier. Deterministic registry code then matches those semantics against catalogue keywords by whole-token overlap — exact token equality, never substring containment, so 'ai' cannot match 'sustainability' — and emits known identifiers. That is why a compiled plan can be audited line by line, why a hallucinated segment cannot reach a buy, and why the same brief compiles the same way twice. **Exposure ledger** — https://adbuymcp.com/glossary/exposure-ledger A single record of advertising exposures across every channel, keyed on whichever of four identifiers a channel can supply — an opaque household key, a hashed advertising id, a consent-based id, or a postcode sector — with per-key legal grading. In practice: The mechanism behind cross-channel frequency deduplication and sequential journeys, and multi-key by design because advertising identifiers are a declining asset: the spine loses precision one step at a time rather than failing at once. Two things are kept strictly separate. The ledger itself never performs an identity join, which happens only in the key-translation service where the ninety-day retention rule is enforced. And its current state — the spine is designed, schema-backed and unit-tested, and is fed only by the deterministic sandbox today. A live report proves aggregate delivery, not household exposure, so the system deliberately refuses to mint identities from one. **Evidence ledger** (also: supply verification status, provenance ledger) — https://adbuymcp.com/glossary/evidence-ledger A six-level grading of how far each supplier integration has actually been proved: sandbox-only, stub, client-built, contract-tested, spend-tested, report-verified. Deliberately not inferred from the presence of code or credentials. In practice: The reason this site can publish a supply matrix that is falsifiable rather than a logo wall. Its governing sentence is worth quoting: a mocked HTTP test proves a client was built, not that a vendor accepted the request, spent money, or returned real delivery. Of twenty-five registered connectors, eighteen are client-built, six are stubs that make no vendor call at all, and one has had read-only calls accepted by a live vendor. Nothing has reached spend-tested or report-verified, and a continuous-integration check keeps those two counts at zero until something genuinely earns them. **Model Context Protocol (MCP)** (also: MCP, MCP server) — https://adbuymcp.com/glossary/mcp An open protocol letting AI assistants call external tools and data through a standard interface, so a model can operate software rather than only describe it. An integration standard, not a capability in itself. In practice: We are named after it, and the name is not the moat — Amazon and other advertising platforms already expose MCP capabilities, and an MCP server alone differentiates nothing. What is unusual here is that the agent transport and the browser exercise the same REST routes, the same tenant boundary, the same wallet gates and the same approval records, so there is no second execution path with weaker rules. The stronger guarantee is what is absent: eleven consequential operations have no agent tool at all, including approval resolution, mandate revocation and issuing a spend grant. An agent prepares and proposes; a named human authorises money. Access today is a self-hosted, single-tenant deployment behind a bearer token — there is no public connector. ## 14. Security, privacy and what is not held — https://adbuymcp.com/security What is enforced in code, what fails closed, what data is held and for how long — and, in its own section, what we do not have yet. The second list is the one worth reading first. Reviewed: 2026-09-01. ### What data AdBuyMCP holds Three categories, and it is worth separating them because they have different risks and different retention. Campaign configuration is yours and is held for as long as your workspace exists. Vendor credentials are encrypted at rest and are the highest-value target here. Exposure and outcome data is the category a DPO will ask about, and it is the one with a hard retention rule. - Campaign and plan data: Personas, compiled targeting specifications, plans, creatives, delivery rows and statements, scoped to your advertiser workspace. - Vendor credentials: Encrypted at rest with AES-256-GCM over the JSON-serialised credential object, keyed on SHA-256 of a master key held in the environment. Never logged, never returned by an API read, and the connection store's health is a startup and readiness check rather than an assumption. - Exposure and outcome events: Ad exposures keyed on an opaque household key, a hashed mobile advertising id, a EUID or a postcode sector, plus outcomes you import or that arrive through your own analytics. Hashed advertising ids are held for at most 90 days. Today the ledger is written to only by the deterministic sandbox — there is no inbound ingest route for viewshed panels, CTV logs or IP logs — so the retention rule is in force before there is live data for it to govern. ### The identity boundary One module in the whole platform performs identity joins, and everything about the privacy posture follows from that. The exposure ledger — which every channel writes into — never resolves a key back to a person or across key spaces. Resolution happens only in the key-translation service, which is where the retention rule is enforced and the only place a reviewer needs to read to understand what is joined to what. - 90-day maximum on hashed advertising ids: After the window, the hashed id is persistently stripped from the exposure event. The event is re-graded to a coarser key rather than deleted, so history is preserved without the identifier. - Multi-key by design: Household key, hashed advertising id, EUID and postcode sector. A spine that depended on advertising ids alone would degrade on a schedule set by Apple and Google; this one loses precision one step at a time and records which step it is on. - Per-key legal grading: Each key space carries its own legal grade, so what may be done with a consent-chained EUID is not what may be done with a postcode sector, and the difference is enforced rather than documented. ### Lawful basis, as an artefact rather than an assurance Every compiled channel plan carries a lawful-basis manifest: the provider of each segment used, the basis it relies on — consent-chained, legitimate interest, or outside scope — and the warnings raised while compiling it. It exists so that the answer to "where did this audience come from?" is a document a DPO can read rather than a conversation. What it is not is a compliance guarantee, and the distinction matters: the manifest attaches conditions you still owe, such as requiring a Legitimate Interests Assessment on file for a given source, or verifying a vendor's consent-framework attestation on connection. It tells you what remains to be done. Turning those conditions into enforced launch gates is work we have not finished. - Outside-scope channels are marked as such: Cinema and DOOH delivery involve no personal data — venue catchments, OpenOOH venue types, postcode polygons and dayparts — and the compiler records that they sit outside UK GDPR scope rather than leaving it to be inferred. Movement data is absent from that list because none reaches the product: the Adsquare contract does not exist. - Warnings travel with the plan: A low-confidence segment match, an unmapped geography or an excluded segment all raise warnings that stay attached to the compiled plan rather than being resolved silently. ### Tenancy and access Every route is scoped to an advertiser workspace and a cross-tenant request returns a 404 rather than a 403, because a 403 confirms that the resource exists. Authentication runs through Clerk organisations mapped to workspace roots. - Actor class cannot be changed by a header: Every request carries a verified actor class. A human web session paired with the agent source marker is rejected rather than silently taking the browser path, and agent credentials must carry a signed agent claim and a verified client id. - Consequential permissions are server-controlled: Exact-digest approval, operator reconciliation and supplier authorisation are three separate permissions issued only from server-controlled roles. An ordinary human tenant member fails closed with a 403, and agents cannot hold any of them. - Audit trail: Actor, source and client attribution are persisted on every action, and the audit log is readable through both the dashboard and the MCP surface. ### What fails closed The useful security question about a young platform is not what it protects but what it refuses to do when something is missing. These are the refusals that are in the code today. - Production will not start without its configuration: Startup fails without Clerk, a strong master key, PostgreSQL, explicit HTTPS origins and a public API asset origin. A legacy plaintext credential row, ciphertext that will not decrypt with the current key, or a missing live credential is a deployment stop rather than a warning. - A launch the wallet cannot fund does not launch: The media spend check is enforced by default; only a sandbox preview may opt into warn-only. Asset production has no warn-only path at all, because a provider cost is incurred at submit. - An ambiguous supplier outcome stops the run: It is never retried and never assumed failed. It raises an exception that requires explicit human resolution, and the operator is handed the vendor, the operation and the idempotency token. - Expired supplier authority raises a spend lock: Before any scheduler work runs. One incident-bound emergency pause is sent, and only a signed human reconciliation clears the stop. - The MCP boundary requires an explicit allowlist: Production must set exact allowed hosts and browser origins — `none` where there are no browser clients. Local execution binds to loopback. Verifier time, request concurrency and JSON body size are bounded, and a request missing its protocol version header is rejected before dispatch. ### How AI is used, and where it is not trusted The AI layer runs on Claude for persona extraction, planning, creative pre-check, optimisation and fidelity narration. What matters for a risk review is the boundary rather than the model. - The model never emits a vendor identifier: It extracts semantics; deterministic registry tables emit segment ids by matching those semantics against catalogue keywords. A hallucinated segment cannot reach a buy. - The model does no arithmetic: Every number in a plan is computed deterministically from adapter forecasts. The model explains the plan; it does not calculate it. - Format limits are schemas, not prompts: Copy that exceeds a channel's limit is rejected at the boundary as a validation error rather than truncated on its way to a provider. - Provenance is visible: AI-produced output is distinguished from deterministic fallback output wherever it appears, and with no API key configured the whole system falls back to those heuristics rather than failing. - Your data is not training data: Briefs and campaign data are sent to the model provider to produce your output and are not used to train models. ### WHAT IS NOT HELD This list is longer than most vendors publish and shorter than the truth would be if it were hidden. None of these is in progress. - ISO 27001 — Not held. No certification, and no audit in progress. This is a young product. - SOC 2 — Not held. No Type I or Type II report exists. - Cyber Essentials — Not held. The most achievable of the three at this stage, and it is still not held. - Penetration test — Not commissioned. No third-party test has been run against this platform. - Public OAuth for multi-tenant MCP — An open launch gate. The server publishes protected-resource metadata and scope challenges, and refuses to mix OAuth with static tokens. What is not done is connecting an authorisation server issuing per-user, per-organisation, audience-bound tokens with PKCE and consent. A static bearer token remains suitable only for a private single-tenant deployment. - A release-evidence PostgreSQL migration set — Partly met. A PostgreSQL migration set is committed and continuous integration runs a smoke job against PostgreSQL 16 on every push, verifying schema drift, running the migration deploy, seeding and smoke-testing. What is not yet release evidence is the newest work: of the 26 migration directories in the tree, 13 are committed, and our own checklist says untracked migration directories cannot serve as release evidence and the rehearsal must be rerun once they are reviewed and committed. The canonical schema remains SQLite for the zero-config sandbox. - ICO registration and a completed DPIA — Open items. Registration for the data protection fee and the exposure-ledger data protection impact assessment are both open decisions rather than completed work. - Independent media-quality verification — Not integrated. No fraud, viewability or brand-safety verification partner is wired in. Placement reporting, supply-path transparency and independent verification are named as unbuilt capabilities. - Support SLA, escalation path or status page — Not built. There is no service level agreement, no uptime commitment, no on-call rota and no status page. A shared support inbox is listed as a pre-launch item that has not been done. - Reviewed legal templates — Templates, pending counsel. The privacy policy, terms, data processing agreement and subprocessor list exist as templates and are marked for review with counsel before they are relied on. - A completed DPIA or LIA for a live campaign — Not yet produced. The lawful-basis machinery produces the evidence a DPIA would draw on, and turning current lawful-basis warnings and manifest conditions into launch gates with completed assessments is named as a pre-launch task rather than a finished one. ### Subprocessors that actually receive data - Anthropic, for The AI layer: persona extraction, plan explanation, creative drafting and pre-check. Data: Campaign briefs and creative prompts. Not used to train models. - Clerk, for Authentication and organisation membership. Data: Account identity for the people in your workspace. - Stripe, for Wallet top-ups. Data: Payment data, held by Stripe rather than by us. - Plausible Analytics, for Cookieless product and website analytics. Data: Page views with no cookie, no cross-site identifier and no IP retention. Four, because those are the four that receive anything today. A longer list exists in our own legal templates and it is wrong in both directions: it names six vendors as active recipients that are fail-closed stubs receiving nothing, and it omits the AI media providers that do receive prompts once asset production is switched on. The per-channel supplier list a live deployment would engage is on the supply page, marked by evidence grade, and no data has flowed to any of them. The full list will be published when it is accurate; in the meantime ask, and you will get the current position rather than a template. ### Where it runs There is no multi-tenant hosted service yet. A deployment today is one you or we stand up: a single virtual machine running the API, dashboard and MCP server behind a reverse proxy, or the three services split across a container host, or the dashboard on a platform host with the API elsewhere. That means data residency is a deployment choice rather than a fixed answer, and for a UK design partner it can be UK-only. The whole money surface is sterling, reporting runs on a Europe/London calendar, and a connected Google customer's timezone must be exactly Europe/London for the current flight contract. ### Insurance Professional indemnity and cyber cover are held by Tenhaw LTD, the operating company, rather than by a product entity. Certificates are available on request during a design-partner conversation. Cover levels are not published here because they are contract-specific. ### What is actually proved, and by what - An acceptance suite provisions a disposable database at a fixed instant, starts the real API and a production build of the dashboard, and runs human-readable scenarios end to end against them rather than against mocks. - A dependency-free browser smoke executes every dashboard page, asserts React hydrated, and fails on an error shell or a severe runtime error. - A coverage gate fails the build unless the line-weighted aggregate across the domain packages, the API and the MCP server reaches 90%. - Regression tests protect the honesty invariants specifically: targeting, tenancy, prepaid commitments, creative readiness, restart-stable delivery, modelled-reach labelling and retention. THE LIMIT OF THAT EVIDENCE: All of it exercises the deterministic sandbox. A green suite proves the software behaves as specified; it proves nothing about a vendor accepting a request or an invoice reconciling, which is what the supply page grades separately. Q: Are you ISO 27001 certified? A: No. Nor SOC 2, nor Cyber Essentials, and no third-party penetration test has been commissioned. ICO registration and the exposure-ledger impact assessment are also open items rather than completed ones. If a certification is a hard procurement requirement, we do not clear it today, and there is no benefit to either side in exploring further. What exists instead is a documented, fail-closed architecture and a published account of what is not held — on this page, rather than discovered in a questionnaire. Q: How long do you keep personal data? A: Hashed mobile advertising ids are held for at most 90 days, after which they are persistently stripped from the exposure event, re-grading it to a coarser key rather than deleting the record. Campaign configuration, plans and delivery data are held for the life of your workspace and deleted on request. Vendor credentials are encrypted at rest with AES-256-GCM and are deleted when the connection is removed. Q: Who can see my campaign data? A: Your advertiser workspace is the scope boundary for every route, and a cross-tenant request returns a 404 rather than a 403, because a 403 confirms that the resource exists. Actor, source and client attribution are persisted on every action and readable in the audit log. The disclosure that belongs beside that: the audit log records what came through the API, not what someone with direct database access could read, and during a design-partner engagement that someone is us. No certification attests to how that access is controlled, because we hold none. Q: Can an AI agent authorise spend on my account? A: No. Agents cannot hold the exact-digest approval permission, cannot resolve human approvals, cannot revoke or replace authority and cannot close missions. Consequential calls fail closed unless the request matches an approval a human granted for those exact bytes, and enabling spend on a staged live line is a separately granted action that approval alone cannot perform. Q: What is your lawful basis for the targeting data? A: It varies by segment, which is exactly why it is recorded per segment rather than asserted once. Every compiled plan carries a manifest naming each segment's provider and its basis: consent-chained where the provider asserts a consent chain, legitimate interest where the segment relies on one, and outside scope for cinema and DOOH, which involve no personal data in delivery at all. Where a segment match is weak or a geography is unmapped, the warning stays attached to the plan. Q: Is my data used to train AI models? A: No. Briefs and campaign data are sent to the model provider to produce your output and are not used for training. The whole AI layer also has a complete deterministic fallback, so the product runs with no model provider configured at all. Q: Where does the data live, and can we keep it in the UK? A: Yes, because there is no multi-tenant hosted service to be in the wrong place. A deployment today is one that is stood up for you — a single virtual machine running the API, dashboard and MCP server behind a reverse proxy, or the three split across a container host — so residency is a deployment decision rather than a fixed answer. Reporting runs on a Europe/London calendar and the money surface is sterling only. Q: Who are your subprocessors? A: Four today: Anthropic for the AI layer, Clerk for authentication, Stripe for wallet top-ups and Plausible for cookieless analytics. That is deliberately shorter than the list in our own legal templates, which is wrong in both directions — it names six vendors as active recipients that are fail-closed stubs receiving nothing, and omits the AI media providers that receive prompts once asset production is enabled. A short accurate list is more use to you than a long inaccurate one. Q: What happens to my money if you fail as a company? A: The prepaid wallet is a real exposure and it deserves a straight answer: funds held in a wallet are a claim on Tenhaw LTD, not client money held in a segregated account. Wallet reservations against live campaigns are held until spend is reconciled to the provider's own invoice, which is designed to limit how far ahead of delivery the balance runs. Be exact about the status of that control: reconciling report rows to a provider invoice is one of the unmet conditions in the launch standard, so it has never once operated. And it would not make the balance bankruptcy-remote if it had. Fund what you are about to spend rather than a quarter in advance, and raise it in the design-partner contract if it matters to your finance team. ## 15. Contact and how to engage — https://adbuymcp.com/contact The one thing worth booking is a 45-minute working session. You bring an audience you actually want to reach and a budget you would actually spend, and it is compiled live into the native targeting each of the seven channels accepts, with the fidelity score, the lawful-basis manifest and the costed plan on screen while you watch. It is not a sales demo and there is no deck: the software is driven in front of you, it runs in a deterministic sandbox that needs no credentials and no card, and if the honest answer is that two of the seven channels deserve your money and five do not, that is what the session tells you. Either way you leave with your own audience compiled seven ways, the scores and the costed plan that came with it, and the sandbox that produced them. ### The three routes in - Book a working session (cal.com/adbuymcp/working-session). 45 minutes. Bring a real brief and we compile it live. The session is the product being driven rather than described. You give an audience in plain English and a budget, and the persona compiler, the seven channel compilers, the plan generator and the fee waterfall all run in front of you. Most of the useful time goes on the two or three fidelity scores that come back lower than you expected, because that is the conversation that decides whether this is worth buying. It is run by the person who built it, not by a salesperson reading from a script. What it is not: It is not a demo of a finished product and it is not a pitch. Nothing is bought on the call, and the compiled plan you see is deterministic sandbox output rather than a media forecast. - Email (hello@adbuymcp.com). The inbox, read by the person who built it. Best for a question with a specific answer: whether a channel can carry a targeting idea, what a fidelity grade means for your audience, what a procurement review can have today, or whether your budget is above the line where we make sense at all. A short question gets a short answer rather than a meeting invitation. What it is not: We have no support SLA, no escalation path and no status page, so this is an inbox rather than a service desk. Say what you need and it gets answered; nobody is on call for it. - GitHub (github.com/Tenhaw). Where the engineering lives. Useful if you are the technical reviewer rather than the person holding the budget. The MCP server, the persona compilers, the adapter provenance ledger that grades every rail, and the fail-closed execution hooks are all in our repository, and the questions worth asking about an agent that can touch a budget are questions about that code. What it is not: Our repository is private today, so this is the organisation rather than a clone-it-yourself link. Ask on the session or by email and the specific file behind any claim on this site can be walked through. ### The working session, minute by minute - 0–5 min — Your brief, in your words. One sentence describing who you want to reach, what you would spend a month, over how long, and what the campaign is for. No qualification questions and no discovery framework. If you would rather send it in advance so the session starts at the interesting part, email it. - 5–15 min — The persona, extracted and edited. The brief becomes a canonical persona: semantics, geography, firmographics, demographics, intent and exclusions, each as an editable chip with its own confidence. Nothing is inferred silently — an age bound you did not state stays absent rather than defaulting — so this is where you correct what the extractor got wrong. - 15–30 min — Seven compiled plans, and the scores. The persona compiles into what each of the seven channels actually accepts, with a fidelity score and a written rationale on each. This is the part of the session that is worth the time: the published bands run from 85 down to 40, and the low ones are where you find out that a channel somebody has sold you cannot express your audience at all. - 30–45 min — What it would cost, and whether to do it. The costed plan, the channel floors, the data CPM uplifts and the fee waterfall, and then the honest read: which two or three channels your budget can fund with enough weight to prove anything, and whether the answer is that this is the wrong product for you today. Bring: - An audience you actually want to reach, described the way you would describe it to a colleague rather than as a targeting spec. - A monthly budget you would genuinely commit, and the flight length you have in mind. A number you made up produces a plan about a campaign that does not exist. - The channels somebody has already sold you on, if any. Those are the ones worth checking a fidelity score against. - Your current measurement setup, if you have one, and whether your conversions carry a geography. It decides whether a causal claim is reachable at your spend at all. Leave with: - The compiled plan from your own brief: seven targeting specifications, seven fidelity scores, seven lawful-basis manifests, and a costed allocation against real channel floors. - Access to the deterministic sandbox, which needs no credentials and no card, so you can drive the same loop yourself afterwards. - A straight answer on whether your budget clears the line where we make sense, including when we do not. WHAT IT IS NOT: It is not a sales demo, not a pitch and not a qualification call. There is no deck, nothing is sold on the call, and there is no follow-up sequence to be enrolled in. What is being recruited is a small number of paid design partners, so the session is as much us working out whether we can serve you as the other way round — and the answer is sometimes no, which is cheaper for both sides established in forty-five minutes than in month three. ### Four things this product does not have - A phone number: We do not have one. Tenhaw LTD, the operating company, publishes its own contact details at https://tenhaw.com, and sending a question about a fidelity score there would reach a consultancy rather than anybody who can answer it. Email or the session, both of which reach the same person. - A support SLA, an escalation path or a status page: None of the three is built, and our own launch checklist lists them as outstanding rather than omitting them. A design partner gets a direct line to the person who wrote the code, which is a different arrangement rather than a better-branded version of the same one. - An account team or a named launch specialist: AdBuyMCP is software rather than a media agency. Nobody here writes your strategy deck, and there is no account management layer to be assigned to. What replaces it is that the fee is published, no undisclosed rebate is taken, and the fee sits on the statement beside the media it was charged on. - A self-serve signup that ends in live media: No buying rail has cleared the launch standard, so real spend runs inside a design-partner engagement rather than behind a signup form. The sandbox, which is the whole product minus real money, needs no conversation at all. ### What a procurement or security reviewer can have today - Legal entity: Tenhaw LTD, company 12735685, incorporated 10 July 2020, registered in England and Wales. AdBuyMCP is a product of that company rather than a separate entity. (https://find-and-update.company-information.service.gov.uk/company/12735685) - VAT: GB388977014 - Registered address: London, England, United Kingdom. - Assurance held: None. No ISO 27001, no SOC 2, no Cyber Essentials, no penetration test, no ICO registration and no completed DPIA. The security page lists all of them as not held rather than as in progress, because none of them is in progress. (https://adbuymcp.com/security) - Legal documents: The privacy policy, terms, data processing agreement and subprocessor list exist as templates and are marked for review with counsel before they are relied on. Ask and you get the current position rather than a document with a review note stripped off it. - Insurance: Professional indemnity and cyber cover are held by the operating company rather than by a product entity. Certificates are available during a design-partner conversation. Cover levels are not published here because they are contract-specific. Q: Is the working session a sales demo? A: No. There is no deck and nothing is sold on the call. You give a real audience and a real budget, and the software compiles them in front of you: seven targeting specifications, seven fidelity scores, a lawful-basis manifest each and a costed plan against real channel floors. The most useful part is usually the channels that score badly for your audience, because that is the part no vendor volunteers. It is run by the person who built it. Q: Do I have to book a call to try it? A: No. The whole loop — brief, persona, seven compiled plans, creative generation and pre-check, approval, launch, delivery, measurement, journeys and optimisation — runs in a deterministic sandbox that needs no credentials and no card, and returns realistic repeatable data rather than a scripted demo. What needs a conversation is real money, because no buying rail has cleared the launch standard yet and live spend runs inside a design-partner engagement. Q: What is your phone number? A: We do not have one, and inventing one for a contact page would be the wrong kind of tidy. Tenhaw LTD, the operating company, publishes its own contact details at tenhaw.com, but that route reaches a consultancy rather than anybody who can answer a question about a compiled targeting specification. Email reaches the person who built it, and the working session is the same person for forty-five minutes. Q: How quickly will you reply? A: There is no service level agreement to quote and it would be dishonest to imply one: we have no support SLA, no on-call rota and no status page, and our own launch checklist says so. What can be said is who reads the inbox, which is the person who wrote the code, and that a specific question gets a specific answer rather than a meeting invitation. Q: What should I bring to the session? A: An audience described the way you would describe it to a colleague, a monthly budget you would genuinely commit, and the flight length you have in mind. If you already know which channels somebody has sold you on, bring those too — they are the ones worth checking a fidelity score against. Sending the brief in advance means the session starts at the compiled output rather than at the typing. Q: Is my budget too small to be worth a conversation? A: Possibly, and it is better to know now. The published band starts at £2,000 a month, but the honest floor is around £10,000: below that a 10% fee is a few hundred pounds against the buyer who needs the most support, and a multi-channel plan cannot carry enough weight on any single line to pace or to prove anything. If your total monthly media is under that, buy search and social directly and come back when there is a second channel worth funding. Booking a session to be told that is still a use of forty-five minutes, and it is the answer you will get. Q: Can I speak to one of your customers? A: No, because there are none. There are no case studies, no named references and no published outcome study, and the design-partner cohort is being recruited now. A reference invented to fill that gap would be worth less than the admission. What can be checked instead is the sandbox, the supply matrix that grades every rail on our own evidence ledger, the fee waterfall and the methodology. Q: We are a procurement or security reviewer. What can we have? A: The register entry — Tenhaw LTD, company 12735685 — the security page's full list of what is not held, the subprocessor position as it actually stands rather than as the template describes it, and the supply matrix. What does not exist is a certification, a penetration test report or a counsel-reviewed contract set, and every one of those is listed as absent at https://adbuymcp.com/security rather than being left for you to discover. Ask for the rest by email and you get the current position. ## 16. The questions asked most often — https://adbuymcp.com/faq The site splits 148 questions across 21 categories, each with its own page and its own FAQPage node, because a single node carrying every question gets truncated by consumers and gives a retriever no usable chunk boundary. The categories: - What AdBuyMCP is (3): https://adbuymcp.com/faq/what-adbuymcp-is - What you can buy today (10): https://adbuymcp.com/faq/can-i-buy-media-today - Fees, wallet and minimums (7): https://adbuymcp.com/faq/what-it-costs - Fidelity scores and lawful basis (6): https://adbuymcp.com/faq/fidelity-scores - Whether this is for you (3): https://adbuymcp.com/faq/is-this-for-me - The four packages, in detail (5): https://adbuymcp.com/faq/the-four-packages - Running a campaign (4): https://adbuymcp.com/faq/running-a-campaign - Television and streaming (4): https://adbuymcp.com/faq/television-and-streaming - Search, social and audio (6): https://adbuymcp.com/faq/search-social-and-audio - Cinema and out-of-home (3): https://adbuymcp.com/faq/cinema-and-out-of-home - Creative and clearance (6): https://adbuymcp.com/faq/creative-and-clearance - Proving it worked (6): https://adbuymcp.com/faq/proving-it-worked - The agent and the MCP server (7): https://adbuymcp.com/faq/the-agent-and-the-mcp-server - Security and data protection (9): https://adbuymcp.com/faq/security-and-data - The alternatives (4): https://adbuymcp.com/faq/the-alternatives - Using the guides (4): https://adbuymcp.com/faq/using-the-guides - Buying the awkward channels (16): https://adbuymcp.com/faq/buying-the-awkward-channels - Measuring it and operating it (15): https://adbuymcp.com/faq/measuring-and-operating - What the words mean (3): https://adbuymcp.com/faq/what-the-words-mean - What each channel costs (14): https://adbuymcp.com/faq/what-each-channel-costs - Doing it, step by step (13): https://adbuymcp.com/faq/doing-it-step-by-step Every one of those questions appears in this file already, under the section that owns its answer. What follows is the set that is about the company and the product as a whole rather than about one page. Q: What is AdBuyMCP? A: A UK paid-media control plane that an AI agent can operate. You describe one audience in plain English and it compiles that into the native targeting each channel actually accepts across seven of them: CTV and streaming, podcast and audio, addressable and linear TV, cinema, digital out-of-home, paid search and paid social. Every compiled audience carries a fidelity score for how close the translation got and a lawful-basis manifest for what made it lawful. It ships an MCP server, so Claude drives the whole loop through the same routes, spend caps and approvals the dashboard uses. It is built and operated from London by Tenhaw LTD. Q: Can I sign up and buy media today? A: You can run the entire product today in a sandbox that needs no credentials and no card. You cannot yet buy media self-serve. No buying rail has cleared the launch standard — a contracted supplier request accepted, a controlled paid activation, immutable creative identity, report rows reconciled to an invoice, and a rehearsed failure — so live spend runs inside a design-partner engagement rather than behind a signup form. The supply page grades all 25 connectors on our own six-level evidence ledger. Q: What does it cost? A: 10% of media spend, disclosed per line, with no spread. Above £25,000 of trailing 30-day spend the rate drops to 5% plus a £749 monthly platform fee. AI asset production is charged separately at a published rate and never blended into the media fee. There is no subscription and no seat fee at the standard rate. What you fund up front is the wallet, prepaid with a £100 minimum top-up, and above the threshold that £749 is a fixed monthly charge rather than a usage one. A per-channel plan floor is enforced when the plan is built. Below about £10,000 a month in total media this is the wrong product, and the pricing page says why. Q: Who is behind it? A: Tenhaw LTD, company number 12735685, incorporated 10 July 2020 and registered in London. AdBuyMCP is a product of that company rather than a separate entity, and the register entry is linked from the footer so it can be checked rather than taken on trust. Q: What is a fidelity score? A: A number out of 100 answering one question: how much of your audience definition survived the translation into this channel's native targeting? It is not a quality score and not a forecast. CTV reaches 85 with a person-level grade when consent-chained B2B segments match strongly. Cinema scores 40 with a venue grade, because a cinema buy is a projection from catchment demographics. Both numbers appear on the plan, in the same size type, before you fund anything. Q: Why publish a score of 40? A: Because a site that only publishes the 85 has told you nothing about the method. Every buyer in this market has been told that a billboard can target IT decision makers, and they already suspect it is not true. Naming the number is the only version of that conversation that leaves you better informed, and it is the reason to believe the 85. Q: Is the AI making the buying decisions? A: The AI extracts semantics from your brief, drafts creative, explains plans and proposes optimisations. It does not compute budgets — deterministic code does all arithmetic, grounded in adapter forecasts — and it never emits a vendor segment identifier, because registry tables do that. Optimiser proposals wait for a human. Consequential actions fail closed unless the request matches an approval a human granted for those exact bytes. Q: Do you have customers I can talk to? A: No. There are no case studies, no named references and no published outcome study, because the design-partner cohort is being recruited now. A testimonial invented to fill that gap would be worth less than the admission. What can be checked instead is the sandbox, the supply matrix, the fee waterfall and the methodology. Q: What is the catch? A: Four, all of them on the site rather than in a footnote. Nothing here has ever spent money through a live rail, so you would be an early partner rather than a customer. It has no security certifications, no penetration test, no support SLA and templated legal documents pending counsel. The fourth measurement verb, brand lift, is a placeholder rather than a feature. And the exposure ledger that powers cross-channel deduplication is fed only by sandbox data today. If any of those is disqualifying, it is better established now than in week three. Q: Why is it called AdBuyMCP? A: Because the MCP server is what changes how the product is used: you describe an audience to Claude and it runs the campaign. It is worth being clear that the name is not the moat. MCP is an integration surface and other advertising platforms expose one too. The defensible part is what sits behind it — one audience compiled into seven native vocabularies with the approximation published — and the name points at the door rather than at the building. ## 17. How to engage Book a working session: https://adbuymcp.com/#book, booked at cal.com/adbuymcp/working-session. 45 minutes. Bring a real brief and we compile it live. It is a working session rather than a sales demo. A real audience and a real budget are compiled live into seven targeting specifications, each with a fidelity score and a lawful-basis manifest, and the reader leaves with the deterministic sandbox whether or not anything comes of it. There is no signup that ends in live media, because no rail has cleared the launch standard. The sandbox needs no credentials and no card. Real spend runs inside a paid design-partner engagement. Contact: hello@adbuymcp.com · https://adbuymcp.com/contact