Maps can now be built by agents over an HTTP API

An agent can now drive a map without a browser. Generate a personal access token in Settings under API access, and the endpoints the canvas uses become available to anything that can make an HTTP request.

The API covers the whole loop. Read a node or a whole subtree to a depth you choose. Create, update and delete nodes with ids you mint yourself, so a client can build a tree in one pass without waiting for the server to name anything. Submit a node and get the finished answer back. Fan a node out one level. Retry one. Interrupt one. The contract is published as OpenAPI, so a client can be generated rather than guessed at.

Two ways in sit on top of it. An MCP server, installable with npx, maps the operations onto tools for any MCP client. An installable skill teaches an agent the same operations in prose.

Submitting a node now waits for its ancestors to finish first, whoever the caller is. That was a real race in the app before this, where a child could be sent with its parent still half written.

Open the app and generate a token under API access in Settings.