Your project management, available to your AI.

What is MCP, in plain English

Out of the box, an AI assistant can only talk. It can draft the perfect task description, but it can't actually put that task in your backlog — it has no hands.

MCP (Model Context Protocol) is the open standard that gives it hands. An app or service exposes an MCP server: a menu of tools it's willing to let an AI use ("create a task", "list members", "post an update"). The AI tool you already use — Claude Code, Cursor, Claude Desktop — acts as the MCP client: it reads the menu, and when you ask for something, it picks the right tool and calls it. Think of it as USB for AI: one connector, and any assistant can plug into any service that speaks it.

Doku ships a hosted MCP server. There's nothing to install or run — it's one HTTPS endpoint, authenticated with your personal token, that works inside the same spaces and permissions you already have:

https://worker.doku.work/api/v1/mcp

What happens when you ask for something

You never call tools yourself — you just talk. The AI decides which tool fits, calls it, and reports back. A real exchange looks like this:

1. You

   "File a bug: login crashes on iOS when the token expires."

2. Your AI

   Reads the tool menu, picks create_task, and fills in the arguments from your message.

3. MCP call

   tools/call → create_task { "title": "Login crashes on iOS…", "project_id": 12 }

4. Doku

   Checks your token and membership, creates the task, and returns it — ✓ Task #1742 created in Bugs.

5. Your AI

   "Filed #1742 in Bugs. Want me to assign it or set a due date?"

Every call is authenticated as you — the AI can only see and touch what your Doku account can. Reads are just reads; writes create real tasks and real notifications, same as if you'd clicked the UI.

Get your token first

In Doku, enable Developer mode in Settings, then use the MCP Link button on any of your groups — it generates a ready-to-paste config with your token already filled in. Every snippet below uses <your-token> as the placeholder. Optional: pin the connection to one workspace with the X-Doku-Space-Id and X-Doku-Project-Id headers.

Connect an AI CLI

Coding CLIs speak MCP over HTTP natively, so this is one command or one config block.

Claude Code — run once in your terminal:

claude mcp add --transport http doku \
  https://worker.doku.work/api/v1/mcp \
  --header "Authorization: Bearer <your-token>"

Then just ask: "create a task in Doku for the bug I just found" — Claude Code discovers the tools automatically.

Cursor and other clients — add this to the client's MCP config (Cursor: .cursor/mcp.json):

{
  "mcpServers": {
    "doku": {
      "type": "http",
      "url": "https://worker.doku.work/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer <your-token>"
      }
    }
  }
}

Connect an AI app

Desktop AI apps like Claude Desktop configure MCP servers with a JSON file. For servers that use a Bearer token, the standard mcp-remote bridge forwards the header for you. Open Settings → Developer → Edit Config (claude_desktop_config.json) and add:

{
  "mcpServers": {
    "doku": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://worker.doku.work/api/v1/mcp",
        "--header",
        "Authorization: Bearer <your-token>"
      ]
    }
  }
}

Restart the app and Doku's tools appear in the tools menu of every conversation — ask for a status summary, file ideas as you brainstorm, or have it post an update that @mentions a teammate. Web-based assistants that only accept OAuth connectors can use the same bridge from a terminal, or drive Doku through the CLI setup above.

What teams do with it

Ask your coding agent to file the bug it just found, break a feature into tasks as it plans, post progress updates with @mentions that actually notify teammates, or pull the team's shared context before making a decision. The agent works inside the same spaces, statuses, and permissions as everyone else — nothing to sync, no second source of truth.