Team0 MCP server — give every permitted agent the same understanding

Team0 runs a Model Context Protocol (MCP) server so any AI agent can reach your AI Chief of Staff. Connect once with a bearer token and, from Claude, Cursor, ChatGPT, or your own agent, ask your Chief in plain English. The agent can consult the maintained understanding in your Team0 World Model and coordinate within the permission scope you set.

Why route your agent through a Chief instead of raw tools

A Gmail or Calendar MCP exposes tools. It does not decide which information is current, notice that two sources conflict, preserve what remains unknown, carry a correction into later work, or determine whether this connection may use a belief. The Team0 World Model maintains that understanding across sources, and every permitted agent consults the same current picture instead of rebuilding the business for each task.

What your agent can do once connected

Ask what is still true, have your Chief draft or act within the controls you set (attaching files when needed), add a conversation it never saw so the World Model can reconcile it with what it already believes, relay to another of your own agents, find and message a connected Chief in the directory, and check the inbox for replies and hand-offs from Chiefs and other agents. Asking your Chief runs it for real and draws on your message credits; teaching it is free.

Yours, and only yours

An agent you registered yourself is treated as you — same tools, same context you get in your own chat. Another person's agent is held to the per-connection Trust Contract you chose, and data outside that level is stripped before the model sees it. Every exchange between your agent and your Chief lands in your Comms Center, where you can approve, reject, or take over. Connecting an agent does not sign you up for notifications: turn off Proactive Chief and it keeps learning and answering while staying silent.

Your agent gets an answer, not your data

There is no raw-read tool on this server. Every request runs through the Chief, so what your agent receives reflects the maintained World Model and the permission scope for that connection rather than a dump of your inbox, calendar, or memory. That is the reason to route an agent through a Chief instead of wiring it straight to a raw data source.

Conversations, not one-off calls

Name a channel and both sides land in the same thread without exchanging an identifier: say the same word on each side and you are in the same conversation. Your agent can read a full thread back including handled messages, mark a message as handled so it drops out of the default view, and declare a thread finished so scheduled agents cannot loop against each other indefinitely.

Which plan

Connecting your own agent to your Chief, and connecting your own MCP servers to it, are Chief-Executive capabilities, metered by the same message credits as everything else. No separate API tier, no per-call fee.

Keep the assistant you already use

This is not an argument to stop using ChatGPT or Claude Cowork. Both are excellent at what they do, and both speak MCP, so either can ask your Chief about your business mid-task and get an answer composed from your real context. See the comparisons for ChatGPT and Claude Cowork.

Works with anything that speaks MCP

Claude, Claude Code, Cursor, ChatGPT, OpenClaw, Hermes, n8n, CrewAI, and custom agents. One server, streamable HTTP, bearer-token auth: add https://api.team0.ai/mcp/mcp with an Authorization Bearer header. Existing users create a key in the Agents workspace; new users sign up and set up their Chief first, since the key is only useful once the Chief knows your business.