MetrqSign in

MCP server reference

Metriq ships a Model Context Protocol server so any MCP-compatible agent — Claude Desktop, Claude Code, or your own runtime — can read your board's flow analytics and drive work items. It is read access (board, items, members, labels, flow summary, forecasts, blockers) plus write access (item CRUD, checklist, labels, epics, and the agent dispatch loop), all team-scoped to a single API key.

  • For the complete reference — setup, the auth model, workflow recipes, the stability promise, telemetry, and the agent end-of-turn contract — see MCP.md in the repository. That document is canon and is the source the public docs derive from.
  • For the full tool list, see the generated MCP tool catalog — never hand-maintained, so it cannot drift from the server.
  • For an agent's working loop — claim a dispatch, drive it to an artifact, hand off for blessing — see the agent skill reference.

Quick orientation

  1. Mint a team-scoped API key in Settings → Agents (admin role).
  2. Point your MCP client at the server, passing METRIQ_API_KEY (and the shared DATABASE_URL / AUTH_SECRET the server validates at boot).
  3. Your agent now sees the tool catalog. The team is derived from the key — no tool takes a teamId, so a call can never reach another team's data.

Host-agnostic by design. Metriq stores artifact URLs as opaque strings and never calls out to the host that produced them. Nothing in the MCP surface assumes a particular runtime, CI system, or git host; any URL your runtime can produce is a valid artifact.