MCP server

Connect Claude, Cursor, or any MCP client. The server exposes the same vocabulary as the UI, CLI, and HTTP API — your agent works the desk as you, with your role, fully audited.

Connect

{
  "mcpServers": {
    "quietdesk": {
      "url": "https://mcp.quietdesk.dev",
      "headers": { "Authorization": "Bearer qd_live_…" }
    }
  }
}

Tools

Query (read — safe by default):snippets.list · snippets.get · cards.list · cards.get · todos.list · milestones.status · activity.list

Operate (write — requires a write-scoped token; clients should confirm):snippets.create · cards.move · cards.comment · todos.add · todos.done

The rules agents live by

1. Attribution. Every write is audited as by <you> (via <client>). There is no anonymous agent action.
2. Gates. cards.move into done on a gated board returns gated_move_requires_human. Your agent can prepare everything; a person approves.
3. Scope. The token's organization and role bound everything — same as the UI.

Example session

You:    What's waiting on my review on the Q3 Roadmap board?
Agent:  → cards.list({ board:"q3-roadmap", status:"in_review", reviewer:"me" })
        Two cards are at your gate: PS-117 (criteria 2/2 ✓) and PS-121 (1/2).

You:    Approve PS-117 and ask Anika to re-check the retry case on PS-121.
Agent:  → cards.move({ card:"PS-117", status:"done" })      ✓ allowed: human caller
        → cards.comment({ card:"PS-121", body:"@anika — retry case…" })
        Done. PS-117 shipped — you approved it, so it cleared the human gate.

Reading these docs by machine? The whole corpus is at /docs/llms.txt and /docs/llms-full.txt.