Connect an AI assistant

This site speaks MCP (the Model Context Protocol), which is how assistants connect to outside tools. You mint a token, paste it into your assistant along with the server address, and the assistant can then look things up here while you talk to it.

We do the looking up. Your assistant does the thinking. We never send your gathering's material to an AI model of our own for this — the excerpts go to whichever assistant you connected, and that company's terms apply to them from there.

Open Account → Identity and find Connect an AI assistant. Name the assistant (“Claude on my laptop”) and press Create token. The token starts with unc_ and is shown once — copy it before you close the box. We keep only a fingerprint of it, so we cannot show it to you again; if you lose it, revoke it and make another.

You can hold up to five at a time — one per assistant, so you can take one away without disturbing the others.

The server address is https://unconference.events/api/mcp.

In Claude, open Settings → Connectors and choose Add custom connector. Give it a name, paste https://unconference.events/api/mcp as the URL, and put your token in the bearer-token / authorization field. Save, then enable the connector in a conversation. Ask it to list your gatherings to check it works.

If your Claude Desktop version wants a config file instead, add a remote server with a proxy such as mcp-remote:

{
  "mcpServers": {
    "unconference": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://unconference.events/api/mcp",
        "--header", "Authorization: Bearer unc_YOUR_TOKEN"
      ]
    }
  }
}

In ChatGPT, connectors are added from Settings → Connectors → create a connector (on plans where custom MCP connectors are available; in the API it is a mcp tool on a response). Use https://unconference.events/api/mcp as the server URL and your unc_… token as the authentication header. Choose “no authentication provider / access token” rather than OAuth: there is no sign-in dance here, just the token.

Add the server to ~/.cursor/mcp.json (or the project's .cursor/mcp.json):

{
  "mcpServers": {
    "unconference": {
      "url": "https://unconference.events/api/mcp",
      "headers": { "Authorization": "Bearer unc_YOUR_TOKEN" }
    }
  }
}

Restart Cursor and the tools appear under MCP in settings.

Exactly what you can see when you are signed in here, and nothing else:

• the gatherings you are a member of — not ones you have never joined, even public ones;
• their schedules and sessions, with hosts by the name and handle they chose;
• session transcripts you are allowed to read. If a gathering keeps transcripts to organizers, and you are not one, your assistant is refused them just as you are;
• search across those transcripts, when the server has a search provider configured;
• the corpus export, if you are an organizer of that gathering.

• anything about a gathering you are not in;
• other people's email addresses, identifiers or private messages;
• votes — neither counts nor who voted, in an open round or a closed one;
• exact addresses. A session someone hosts at their own place is described by the coarse public place they chose; the exact spot, and any private group link, stay behind the session page;
• anything at all that changes: the connection is read-only. It cannot post, edit, RSVP or delete.

A token is not a password. It cannot sign in as you, cannot reach your account settings, and cannot publish anything to your repository on the network.

Go back to Account → Identity, find the assistant in the list, and press Revoke. It stops working immediately — no waiting, no propagation. The list also shows when each token was last used, so you can tell which ones are still doing anything.

Do the same if you think a token leaked. Revoke it, then mint a fresh one and paste that into the assistant.

• 401 — the token is missing, mistyped or revoked. It must be sent as Authorization: Bearer unc_….
• 429 — more than 120 requests in a minute from one token. Wait a minute.
• Search says it is not available — transcript search is switched off on this server, or the transcripts have not been indexed yet. That is an operator setting, not something you can fix; ask the organizers. Listing sessions and reading transcripts in full still work.