What is an MCP task manager, and who needs one?

Short answer: An MCP task manager is a to-do app that speaks the Model Context Protocol (MCP), the open standard AI assistants use to call external tools. After an OAuth sign-in, your assistant reads your tasks, adds new ones, edits them, and marks them done. Done Bear is one: Claude, ChatGPT, Cursor, and any other MCP client connect at https://mcp.donebear.com/mcp.

Most “AI plus to-do list” workflows are transcription. The assistant suggests a plan, and you copy it into your task manager by hand. An MCP task manager removes that step. The Model Context Protocol is the open standard, introduced by Anthropic and now supported by Claude, ChatGPT, Cursor and other clients, that lets an assistant use your tools directly.

When your task manager ships an MCP server, your assistant can act on your real list: triage the Inbox, plan your day around what’s in it, and file follow-ups the moment a call ends. It’s the list you already keep, reachable from whichever assistant you prefer, over a connection you can revoke.

MCP task manager vs. AI coding-task tools

Task Master and Shrimp Task Manager get described as “MCP task managers” too, but they’re coding-task planners. They help an agent break a software feature into sub-tasks and track them for the duration of a coding session. They live in your editor, they’re good at shipping code, and they’re done when the work is.

Your dentist appointment doesn’t go there. Neither do “call the plumber,” “review the Q3 budget,” or “buy a birthday present.” Those belong in your real to-do list, the one you check on your phone, share with a partner, and carry across weeks.

Done Bear is an MCP task manager for that list. It’s a GTD-native, local-first task manager first, with the five views (Inbox, Today, Upcoming, Anytime, Someday) and a Logbook. The MCP server is how your assistant helps you run it.

What you can do with one

Once your assistant is connected to Done Bear’s MCP server, you can ask for things like:

• “Look at my Done Bear inbox and suggest where each task belongs.”
• “Plan my Wednesday: pick five tasks from Anytime that fit around my meetings.”
• “I just got off a call, add follow-ups: send Sam the proposal, book the venue, chase the invoice.”
• “What did I finish this week?”

Every action runs through OAuth. The assistant acts as you, only after you’ve signed in, and only while the connection is enabled. Access is scoped to your workspace, and a read-only token can’t call a write tool.

How the connection works

Done Bear hosts its MCP server over Streamable HTTP with OAuth 2.1, so there’s no local package to install and no API key to manage. Setup depends on the client. With Claude and ChatGPT you add a custom connector and paste the URL. Cursor takes a one-click deeplink, or a two-line mcp.json entry.

Claude Code is one claude mcp add command, and Windsurf wants the server added with serverUrl in your MCP config.

The full menu is on the MCP hub. For the client setup matrix, the tool inventory, and how auth and scopes behave, see the free task manager MCP resources.

Why GTD-native and local-first matter here

Any task manager can add an MCP endpoint. The structure underneath decides how well an assistant uses it.

The five GTD views give it fixed meanings: “move this to Someday” means something specific, so it files things instead of inventing a tagging scheme. And because Done Bear is local-first, the changes it makes show up on your devices straight away, next to the edits you made offline.

Todoist and TickTick ship MCP servers as well, and Things 3 has no official one. Done Bear pairs agent access with GTD-native views and local-first sync, and MCP access is on the free plan.

Done Bear is free to start. Connect your assistant and let it help you run the list you already keep.