procedure
Connect Claude to your AdBuyMCP account
We run the server, and your client connects with your account's access token.
- steps
- 5
- things needed first
- 3
A short application: five questions, no card. Approved accounts get the whole loop in their own sandbox, up to a monthly cap. Live spend is open to paid design partners.
before you start
What you need first
- An approved AdBuyMCP account, which runs the whole loop in the sandbox with no card and no live media
- Your deployment's MCP address and the access token your account holds
- Claude Code, Codex or another MCP client that lets you set a request header
the procedure
5 steps
Every step comes with a note on what commonly goes wrong there.
- step 01
Find your address and token
Your account's deployment has its own MCP address, ending in /mcp, and your account holds the access token for it. The token travels as a bearer token in the Authorization header.
what goes wrong hereTreat the token like a password. Anyone holding it can call your account's tools, so keep it in an environment variable or a secrets manager and out of shared config files and repositories.
- step 02
Add it to Claude Code
Register the address with Claude Code as an HTTP server and pass the token as the Authorization header. Claude Code takes both in one command.
what goes wrong hereThe connector screen on claude.ai signs in with OAuth, which has not shipped here, so it has nothing to connect to yet. Claude Code is the route from Claude today.
- step 03
Or add it to another client
Codex, Cursor, Zed and other MCP clients take the same address and the same header. The transport is stateless streamable HTTP, so the client needs no sticky session with one server.
what goes wrong hereA client that offers only a sign-in button for remote servers is expecting OAuth. Pick one that lets you set a header.
- step 04
Check the connection
Ask your client to list AdBuyMCP's tools. The build has 100, and a deployment can narrow the list to what your account is entitled to use.
what goes wrong hereThe deployment's /healthz endpoint answers without a token. If it answers and tool calls fail, check the token first.
- step 05
Set what the agent may do
Tool calls run under your account's approvals and spend caps, the same as the dashboard. No tool exists for 11 consequential operations, so a named person performs those.
what goes wrong hereA mandate sets the actions, the currency, the money ceiling and the expiry for an agent. Granting and revoking one has its own procedure on this site.
Your MCP client connected to your account's own deployment, calling up to 100 tools over the same routes, tenant boundary and approval gates the dashboard uses.
other procedures
What else you might need to do
- Write a brief an agent can compile
- Read a fidelity score
- Run a geo-lift test
- Check a lawful-basis manifest
- Fund a campaign
- Grant and revoke an agent mandate
Bring a real brief to a 45-minute session and we’ll plan it live.
Questions
Do I need to host or run anything?
No. AdBuyMCP runs the MCP server, and your approved account gets its own single-tenant deployment. The account runs in the sandbox, with no card and no live media. Your side is the MCP client you already use and the access token your account holds. One deployment serves one account, because shared sign-in for many companies is not built yet.
Can I add AdBuyMCP as a connector on claude.ai?
Not yet. A claude.ai custom connector signs in with OAuth, and the authorisation server that would issue per-user, per-organisation tokens has not shipped, so claude.ai has nothing to connect to. Today you connect from Claude Code, Codex or another client that takes your deployment's address and your account's access token. There is no public AdBuyMCP connector and no registry listing.