MCP (Model Context Protocol) is how AI assistants talk to external tools. Fivo runs an MCP server, so any MCP client can read your Brain: Claude, ChatGPT, Cursor, VS Code, Claude Code, Gemini CLI, Copilot and dozens of agent builders.
Once connected, your assistant asks Fivo for context instead of you pasting documents and threads. It gets ready-made facts with sources, in a few hundred tokens rather than thousands.
What you need
- A Fivo account on a paid plan or in its trial. MCP is a paid feature; the 14-day trial unlocks it.
- The MCP server URL from your Fivo settings. Open Settings, then MCP, and copy the URL. The same page has a step-by-step guide for each client.
How clients sign in
Almost every client signs in with OAuth. You add the Fivo server URL, the client registers itself with Fivo, and a Fivo page opens in your browser. You check which app is asking, where it will return to and what it can read, then click Allow. You never paste a password, a key or a token, and you can disconnect the app at any time from the MCP settings page.
Clients that cannot open a browser, and agents you build yourself, use a personal access token instead. You create it on the MCP page, name it, set an expiry and choose its read permissions. Fivo shows it once. It acts as you until it expires or you revoke it; use one token per agent.
Client notes
- Claude (claude.ai, desktop and mobile): add a custom connector with the Fivo URL and leave the OAuth client fields empty. On Team and Enterprise plans an owner adds the connector once, and each member connects with their own Fivo account. See Connect Claude to your Brain with MCP.
- Claude Code: add the server as an HTTP MCP server with the URL; it opens the browser for approval.
- ChatGPT: turn on Developer mode in advanced settings and create a connector with the URL and OAuth. In Business and Enterprise workspaces an admin has to allow Developer mode first.
- Cursor and VS Code: add the server to your MCP configuration with just its URL. The editor registers itself and opens your browser.
- Your own agent: send the personal access token as a Bearer header.
What your AI can do
The Fivo MCP server is read-only. Its tools let an assistant search the Brain, open any fact or entity, list people and companies, get a 360º brief, list commitments, read what happened in a time window, prepare a meeting, read reports and tasks, check your calendar and list your team. It also exposes ready-made prompts for meeting preparation, a weekly digest, an account review, a one-on-one and a decision log. See Questions your AI can answer.
Everything the assistant reads is filtered by your permissions. It sees exactly what you would see in Fivo, no more. Each fact comes with its source and a link to the original tool, so the assistant can cite it or follow up there with its own tools.
Permissions you grant
When you approve a client, you grant read scopes: the Brain, reports, tasks, calendar and organization members. The consent page lists them. You can grant fewer when creating a personal access token. An organization can also restrict which clients are allowed and set a maximum lifetime for tokens.
Status and notices
If a graph is still building, frozen or pending payment, the assistant is told so in the response, so an empty answer is not mistaken for "nothing happened". In the last days of a trial it also warns that the graph will be frozen. The plan tool tells your assistant what plan the workspace is on and, for the owner, how to pay.
Managing connections
The MCP settings page lists connected apps and when each last read your graph. Disconnect an app there, or revoke a token, and it stops on the next request.
- mcp
- claude
- chatgpt
- cursor
- setup