Team0 · Model Context Protocol
Give every agent the same understanding.
Keep the agent you already built. Connect it to your Team0 Chief and it can consult the Team0 World Model: a maintained understanding of what is current, what changed, what conflicts, and what remains unknown. Each connection stays inside the permissions you set.
{
"mcpServers": {
"team0": {
"url": "https://api.team0.ai/mcp/mcp",
"headers": {
"Authorization": "Bearer t0c2c_••••••"
}
}
}
}On Claude Desktop, Claude.ai, or ChatGPT? Skip the file — those apps take a link, not a config. Get your key and paste the personal MCP URL into the app’s connector settings.
› “Ask my Chief what’s slipping this week and draft the follow-ups.”
Tools are not understanding
Connecting tools is easy. Maintaining understanding is the real system.
An MCP connection can expose an inbox or calendar. It does not decide whether a fact is current, notice that two sources conflict, carry a correction forward, or determine whether this agent may use it. That is the work of the Team0 World Model.
The connection layer
Expose the tools.
Connect an inbox and calendar, store past messages, and give an agent actions it can call. Useful, but each session still has to reconstruct what your business means.
The maintained understanding layer
- 01Resolve who and what each signal refers to across inboxes, calendars, meetings, documents, and agents.
- 02Distinguish current beliefs from changed, conflicting, and still-unknown information instead of treating every stored sentence as true.
- 03Keep every belief traceable to its evidence, then retire the old version when a correction or stronger source arrives.
- 04Carry one correction into later answers, briefings, and actions so each agent does not repeat the same mistake.
- 05Apply a separate permission scope before another person’s Chief or outside agent can consult the understanding.
Every permitted agent can consult the same maintained understanding without rebuilding your business from scratch.
Just need one calendar event? Use a calendar MCP. Team0 earns its place when an agent needs current context, grounded judgment, corrections that carry forward, and permissions that travel with the connection.
What your AI can do once connected
Five things, in plain English.
Ask what is still true
Your agent consults the maintained understanding behind your Chief: current commitments, changed plans, unresolved conflicts, and the receipts behind them.
message_chiefWork from the same context
Draft the follow-up, chase the invoice, prep the meeting, or attach a document. The work begins from the same current understanding you see in Team0.
message_chief · attachmentsAdd what it missed
Hand your Chief a conversation it never saw, such as a product session or support thread. The World Model absorbs what matters, reconciles it with what it already believes, and keeps the source.
ingest_episodeCoordinate with permitted agents
Relay to another of your agents, or find and message a connected person’s Chief. Each exchange stays within the Trust Contract for that connection.
message_agent · search_directorySee what’s waiting
Check the inbox for replies and hand-offs from Chiefs and other agents, so nothing stalls mid-conversation.
check_inbox · check_responseAsking your Chief runs it for real, so it draws on your message credits — 1 to 20 depending on the work, tiered the same way your own chat is. Teaching it (ingest_episode) is free. Answers usually land inside 45 seconds; longer jobs hand back a handle you poll with check_response.
Tools vs understanding
Most MCP servers expose actions. Team0 gives your agent maintained understanding.
A generic MCP server lets an agent fetch or post. Team0 routes the request through your Chief and the World Model underneath it: what is current, what changed, what conflicts, and what this connection is permitted to use. The agent works from your business as it stands now, not a context bundle rebuilt for each task.
One maintained understanding. A separate permission scope for every connected agent.
How it works
Three steps. One maintained understanding.
Get your key
Agents → Connect an agent. Scoped to you, revocable and rotatable anytime.
Add one line
Paste the config into your MCP client — Claude, Cursor, OpenClaw, whatever you run.
Talk in plain English
Ask your Chief anything, or tell it to get something done. It takes it from there.
Works with
Anything that speaks MCP.
One server, streamable HTTP, bearer-token auth. If your agent supports remote MCP, it connects.
Permissioned by design
The understanding stays yours. Access is granted connection by connection.
Your own agent reaches what you reach
A key you register for your own agent represents you. It can consult the same Team0 context and tools available in your own chat, through a revocable connection.
Every peer gets a separate scope
A peer’s agent is held to the Trust Contract you chose for that connection. Context outside that scope is removed before the exchange reaches the model.
Your agent gets an answer, not a raw dump
There is no raw-read tool on this server. Every request runs through the Chief, so the response reflects the World Model and the permission scope for that connection.
Every exchange is visible
Everything your agent and your Chief say to each other lands in your Comms Center, alongside the Chief’s email and its conversations with other Chiefs. Approve, reject, or take over from there.
Quiet, if that’s what you want
Connecting an agent doesn’t sign you up for notifications. Turn off Proactive Chief and it stops reaching out entirely — it still learns, still answers when your agent asks, and says nothing until you want it to.
Put every permitted agent on the same page.
One server · api.team0.ai/mcp/mcp