Connect your AI (MCP)
Give your AI assistant a key: it will read and, if you want, edit only the map it is tied to.
- Open the map and press MCP in the top bar.
- Create a key: “read and write” or “read only”. Copy the command we show you (the key is shown only once).
- Paste it in your MCP client (for Claude Code it’s a
claude mcp add …command) and ask the assistant to describe your map.
How your map is protected
- Each key is tied to one map: the assistant can’t name another map, and can’t list them.
- Only the key’s hash is stored. Deleting it from the list revokes it immediately.
- A read-only key doesn’t even have the write tools.
- Keys expire: after 90 days by default (30 days or 1 year to choose; “never” is possible but not recommended). Once expired the assistant gets “key expired” and you create a new one from the MCP panel. Keys created before this feature don’t expire.
- Limits: 120 requests per minute per key and up to 200 nodes per map, within the plan’s limit (15 on free). An assistant can only edit a node’s label, type, group, provider, position, reachability and check, never the status the system writes.
Supported clients
| Client | Status |
|---|---|
| Claude (Code and Desktop) | supported — native header or via mcp-remote |
| Cursor | supported — native, in .cursor/mcp.json |
| GitHub Copilot (VS Code and CLI) | supported — in .vscode/mcp.json; if it tries OAuth on its own, use “/mcp enable” |
| ChatGPT | not yet — its connectors require OAuth, which we don’t have yet |
If many of you use it
Each person has their own map and creates their own key: one user’s assistant can’t read or edit another user’s map. A key works for a single map; with several maps you create one key each. Only a member of the map can create its keys.
If the assistant reads untrusted content (web pages, email), malicious text can trick it into editing your map. In that case use a read-only key.