MCP server Pro
Let Claude, Cursor or Codex read your architecture directly.
Instead of pasting a diagram into a prompt and hoping it's current, your agent asks Canopy. It gets your real services, what they cost, and how they connect - live, every time it asks.
It's a remote server: a URL you paste into your client. Nothing to install, nothing to run.
Agent
Claude · Cursor · Codex
CanopyYour maps
live architecture
Connect
Pick whichever your client supports. Both end up in the same place.
Sign in once in your browser. No token to copy, rotate, or leak.
ReadPaste a workspace token into a header. Works with every client.
ReadBoth use the same address:
https://canopy.8starlabs.com/api/mcpWhich one should I use?
Use OAuth if your client offers it - Claude, Cursor and Codex all do. You sign in, tick what the agent may see, and you can take it back later without disturbing anyone else.
Use a token for CI, scripts, or any client that can't open a browser.
Then what?
The eight tools, the one that draws your map in the chat, and prompts to paste.
ReadAlready have Claude open in your repo? Let it create the map for you.
ReadWhat an agent may see, who can grant it, and how to revoke it.
ReadWhy bother
A diagram in a prompt is a photograph - true when it was taken. Ask an agent "what breaks if the API goes down?" against a pasted diagram and it answers from whatever was true that day.
Connected over MCP, it calls find_dependencies and answers from the map as it
is right now.
Coming soon
Agent-native payments: x402 & MPP
Support for x402 and MPP (machine protocol payments) is on the roadmap, so agents can transact programmatically (e.g. pay for a template or metered access) without a human in the loop. Not available yet, this is a heads-up, not a live feature.