Build a map from your agent Pro

You don't have to build maps in Canopy. If you already have Claude or Codex open in your repo, connect Canopy as an MCP server and ask:

Your stack

A repo, a compose file, a list of tools

Whatever your agent can already read.

Your AI client

create_map

Canopy

A living map

Services, connections, spend - editable in the app.

Needs the maps:write permission. Read-only connections can't create anything.

Your agent already understands your stack, so it does the mapping. It calls create_map with the services and connections it found, and the map appears in your workspace ready to edit. This is a fourth way in alongside importing a repo, pasting canopy.json, and cloning a template - useful when the source of truth is something Canopy can't scan for you, like an internal wiki, a Terraform state, or a whiteboard photo your agent just read.

Two things to know:

• It needs maps:write. Tick it at consent (or include it when you create a token). A connection without it can read your maps but can't create one.
• update_map replaces arrays wholesale. If you send nodes, that becomes the entire node list. Have the agent read the map first and send the full set, or it will drop whatever it left out.