Integrations
MCP
Automate ClusterDeck diagrams from external tools via Model Context Protocol.
ClusterDeck exposes an MCP (Model Context Protocol) surface so assistants and automation can list, read, and update diagrams without clicking through the UI.
What you can do
Typical MCP capabilities (when enabled for your project):
- List diagrams — discover diagram ids and titles
- Get diagram — fetch a full snapshot (
nodes,edges,meta) - Update diagram — upsert/delete nodes and edges; update meta (tags, flows)
- Set / list node status — push or read observed live health (TTL overlay; see Live status)
Use this for:
- Generating or patching architecture from a coding agent
- Keeping diagrams in sync with infra-as-code reviews
- Bulk retitling, tagging, or wiring documented connectors
- Pushing probe results from n8n / cron without rewriting diagram history
Safety rules
- Prefer read (
list/get) before large updates. - Send only the fields you intend to change; understand that arrays you send (e.g.
properties,tags,connectors) replace the previous array. - Do not invent
childDiagramId— nesting is managed by the product when users open/create child diagrams. - Never push secrets into node properties via automation.
Suggested agent workflow
list_diagrams— find the target idget_diagram— read current nodes/edges- Plan the delta (new nodes, edge rewires, property updates)
update_diagram— apply upserts/deletesget_diagramagain — verify
For nested systems: get the parent, then get/update the child diagram by id.
Enabling access
- In the workspace account menu, open MCP token.
- Generate a token and copy it once — store it in your MCP client config.
- Enable MCP per project (toggles in the same dialog or in the editor).
- Point the client at the public MCP URL (for example
https://your-host/mcp) with:Authorization: Bearer <token>.
Only projects with MCP enabled appear in list_diagrams. Revoke the token from the same dialog if it leaks.
While an agent edits, the editor may show an MCP agent editing indicator.
Related
- Nested diagrams — how trees are modeled
- Export & import — offline bundles vs live MCP updates
- Nodes / Edges — field meanings agents must respect