How to Let Claude Manage Your To-Do List (MCP Tutorial)

TL;DR: Done Bear runs a hosted MCP server at https://mcp.donebear.com/mcp. Add that one URL to Claude Desktop, Claude Code, Cursor, VS Code, or any MCP client, sign in once via OAuth, and your assistant can list, create, edit, and complete your tasks. No API keys, no local install.

Most “AI + to-do list” workflows are copy-paste: the assistant drafts a plan, you retype it into your task manager. The Model Context Protocol (MCP) removes that gap. It’s the open standard that lets AI assistants use your tools directly, with your permission. Here’s the whole setup.

What can an AI assistant do with my tasks?

Once connected, your assistant can read a workspace snapshot, list and search tasks, create and edit them, mark them done, and manage projects and labels. Every action goes through OAuth: the assistant acts as you, only after you’ve signed in, and you can revoke access at any time.

In practice, you can ask:

• “Look at my Done Bear inbox and suggest where each task belongs.”
• “Plan my Wednesday: pick at most five tasks from Anytime that fit around my meetings.”
• “I just got out of a call. Add follow-ups: send the proposal to Sam, book the venue, chase the invoice.”
• “What did I actually finish this week?”
• “Create a project called ‘Kitchen renovation’ and break it into first steps.”

Step 1: Create a Done Bear account

Sign up free at donebear.com. The free plan (up to 250 tasks) covers this whole workflow. Add a few real tasks so your assistant has something to work with.

Step 2: Add the server to your client

The endpoint is the same for every client:

https://mcp.donebear.com/mcp

Claude Desktop: Settings → Connectors → add a custom connector, paste the URL.

Claude Code:

claude mcp add --transport http donebear https://mcp.donebear.com/mcp

Cursor: add to .cursor/mcp.json (or ~/.cursor/mcp.json globally):

{
  "mcpServers": {
    "donebear": { "url": "https://mcp.donebear.com/mcp" }
  }
}

VS Code (Copilot): add to .vscode/mcp.json:

{
  "servers": {
    "donebear": { "type": "http", "url": "https://mcp.donebear.com/mcp" }
  }
}

Any client that speaks MCP’s streamable HTTP transport works the same way. The client-by-client guide (including Codex and Windsurf) is in the MCP setup docs.

Step 3: Sign in once

Restart your client and ask: “List my Done Bear tasks.” The first request opens a browser window to sign in; after that, authentication is automatic.

A daily workflow that sticks

Morning triage (2 minutes). “Go through my Done Bear inbox. For each task, tell me whether it’s Today, Upcoming, Anytime, or Someday, and why. Then apply it.” You veto, it files.

After-call capture. Paste your raw meeting notes and say “extract the action items and add them to my Done Bear inbox.”

Weekly review. “Compare my Logbook to what’s still in Anytime. What am I avoiding?” An assistant with read access to your real list gives you a review no motivational blog post can.

Why this works best in a GTD-native, local-first app

The five GTD views give your assistant real semantics: “move this to Someday” means something specific, so it files things instead of inventing a tagging convention. And local-first sync means its changes appear on your devices in real time, alongside edits you make offline.

Todoist and TickTick ship hosted MCP servers too (TickTick’s is in beta, on paid plans), and Things 3 has none. What no competitor pairs with agent access is GTD-native views plus local-first sync, on a free plan.

Prefer the terminal? The same capabilities are in the CLI.

Start free, connect the server, and let your assistant do the filing.