Agent access — connect Claude or another AI client
What it is
Your account can answer an AI agent directly, over MCP (Model Context Protocol) - the way Claude and other AI clients connect to outside tools. Ask it in plain language: what are my open gaps, which topics am I measured on but never named for, who is being cited instead of me. It is read-only. Ten tools read your visibility runs, the sources engines cited, your gap list, the evidence you have published and your entity - and nothing an agent can call publishes, approves, edits or deletes anything. The address, for every way of connecting, is exactly: https://therankingfactory.com/api/mcp Use it exactly as written. A client that signs in checks it against what our server says it is, so a different spelling of the same address can fail.
Connect Claude by signing in (Agency and Enterprise)
1. In Claude, add a custom connector - in Claude's own words at the time of writing, Customize, then Connectors, then Add custom connector - and enter the address above. 2. Claude sends you here. Sign in if you are not already. 3. The page shows who is asking, where you will be sent back to, and what it may read. For Claude, who is asking is claude.ai: that is the address that vouches for it, rather than the name it gives itself. 4. Choose what it reads: every project in the account (including ones you add later), or one project only. 5. Allow. Claude finishes connecting, and there is no token to copy or keep safe. A connection given one project reads that project and no other, whatever it is asked: every tool answers for that project, and asking about another is refused. That is the choice to make when you connect on a client's behalf. Clients that run on your own computer, such as Claude Code, work the same way; the page adds a warning for them, because any program on the machine could claim to be one. Allow it only if you started the connection yourself.
Connect with an agent key (Growth and up)
For any client that cannot sign in, and on plans without sign-in: 1. Settings → Agent access → Create an agent token. 2. Name it for where it will live, choose what it may read, and copy it. It is shown once; we keep only a hash. 3. Give the client the address above and the header: Authorization: Bearer rft_… A key sees every project in the account. Ask any tool for one project by passing its domain or name.
Seeing and ending connections
Settings → Agent access lists every application connected by signing in - what it reads and when it was last used - with Disconnect beside each, and every key with Revoke. Either stops it at once. A signed-in connection also stops by itself when the account moves off Agency or Enterprise (within the hour), and if an old copy of its sign-in is ever presented again, which we treat as a sign it was copied. Connecting again takes a minute, and nothing about your account is lost.
Every number an agent gets comes with the sample it was measured over, and says which project it covers. If an agent quotes a rate to you without either, ask it for them.