MCP tools Pro

What an agent can actually do once it's connected: the eight tools, the one that draws your map in the chat, and prompts you can paste to try them.

Tools

Once connected, agents get these tools (each takes an optional map id to scope to a single map). Every MCP tool only reads - there is no write tool, so nothing an agent does over MCP can change a map. Which tools appear depends on the permissions you granted:

Tool	Returns
`get_architecture`	a map's full nodes + edges (or the map list)
`list_services`	every node with its tech, cost and billing
`get_spend`	cost rolled up per map and across the workspace, grouped by each node's currency
`find_dependencies`	upstream/downstream blast radius for a node
`preview_map`	the map drawn, as an interactive diagram in the chat
`create_map`	build a new map from services and connections
`update_map`	change an existing map
`delete_map`	remove a map (archived instead if it has buyers)

The first five only read. The last three need the maps:write permission, so a read-only connection never even sees them listed.

See the map, not just the JSON

preview_map renders the map as a real diagram inside your client, instead of printing nodes and edges as JSON for you to read. Ask "show me my production map" and the map appears in the conversation:

Laid out like Canopy. Nodes you positioned in the editor keep their coordinates; a map built from the API or a pasted canopy.json is auto-arranged with the same layout the editor's auto-arrange uses.
Click a service to trace its blast radius: everything upstream and downstream lights up, the rest dims, and a panel lists both sides.
Pan, zoom, and fullscreen, plus Open in Canopy to jump to the real editor.
Switch maps from the dropdown without going back to the agent.
Matches your theme, light or dark.

How it works

This uses MCP Apps (SEP-1865), the official MCP extension for interactive UI. The tool points at a ui:// resource that Canopy serves; your client renders it in a sandboxed iframe. The view is a single self-contained document with no external scripts, fonts, or network calls, so it only ever sees the map data your client hands it.

Support varies by client. In a client without MCP Apps support, preview_map still works: it returns a short text summary instead of a drawing, and every other tool is unaffected. Check the MCP client support matrix for where the diagram renders today.

Example prompts

Ask your MCP-connected agent questions like these. Each one is copyable:

Show me the architecture maps in this Canopy workspace, then summarize the services and connections in the most recently updated map.

List every service and workflow step in my Canopy workspace with its tool, monthly cost, and billing model.

Which services cost the most each month, and which map do they belong to?

Find the blast radius for the api node. What depends on it, and what does it depend on?

If the database node goes down, which downstream services or workflows should I check first?

Compare all maps in this workspace and tell me which one has the most services, the most connections, and the highest monthly cost.

Show me my Canopy production map as a diagram.

Draw the most recently updated map in my Canopy workspace so I can see how the services connect.

Use node ids

find_dependencies needs a node id, such as api, web, or db. If you only know the display name, ask the agent to call list_services first and choose the matching id.

PreviousMCPMCP Next Build a map from your agentMCP

to paste into any AI.

Ask ChatGPT about this page.

Ask Claude about this page.

Need help?

Have ideas?

General inquiry? Email support.