Todoist MCP Server

The Todoist MCP Server connects your Todoist task list to AI assistants. MCP, the Model Context Protocol, is an open standard that lets assistants such as Claude, ChatGPT, Codex or Cursor reach external services. The server is built by the Todoist team itself, and the source code lives in the Doist GitHub account under the MIT license. You have two options: use the hosted service at ai.todoist.net/mcp, or run the server locally on your own machine. The hosted option needs no installation. Sign-in there uses OAuth. OAuth is a sign-in method where you grant the assistant access in your browser without handing over your password.

What is the Todoist MCP Server good for?

A task list only works if things get into it quickly and get cleaned up regularly. That is where an assistant helps. You can tell it in chat to turn the next steps from a meeting summary into tasks in a specific project. You can ask what is due today and tomorrow, push overdue tasks to next week, or have it set up a project with sections and labels. Developers get a lot out of it too, because Claude Code or Codex can file loose ends from coding work straight into Todoist. No TODO gets lost in the terminal.

Tasks, projects and structure

According to the repository, the server covers everyday work: add, update, complete, reopen and reschedule tasks, and find tasks by date or search term. On top of that come tools for projects, sections, labels, filters, comments and reminders. In shared projects, the assistant can assign tasks and list collaborators. Projects can be exported as templates and imported again. The Todoist team describes its goal like this: a small set of tools that cover whole workflows rather than many tiny atomic actions. That saves the language model extra steps.

Insights and overviews

Besides editing, the server can also report. There are tools for a daily overview, productivity stats, project activity and project health. For teams, workspace insights come on top. That lets you ask which projects have stalled or what got done last week. The server also offers search and fetch tools that follow OpenAI's MCP specification, which matters for ChatGPT. Through MCP Apps, it can show small widgets such as task lists right inside the chat in clients that support them.

Hosted or local?

For most people, the hosted service is the easiest route. In Claude, Todoist is listed in the connector directory; in ChatGPT, it is a plugin that then also shows up in Codex. For Claude Code, there is an official plugin that sets up the server with two commands. Cursor and other clients connect to the URL through the mcp-remote helper, and VS Code connects directly over HTTP. If you prefer to work locally, start the server with npx and a Todoist API token. You will find the token in Todoist's settings under Integrations. The repository gives another reason to run it locally: the hosted service can drop sessions frequently. A local HTTP server is the workaround. By default it only listens on your own machine and checks incoming requests against a list of allowed host names.

How does it differ from alternatives?

There are several unofficial community Todoist MCP servers. Most of them only run locally and always need an API token. The official server is maintained by the Todoist team, works both hosted and locally, and is available as a connector in the major assistants. One detail stands out: the tools are not tied to MCP. Developers can pull them into their own apps as an npm package, for example with Vercel's AI SDK.

Know the limits

Todoist's help page for the Claude connector lists a few limits. The assistant handles at most 25 tasks per action. If it changes the date of a recurring task, the recurrence is lost, so ask it to reschedule instead. Depending on the client, some features may be missing, such as managing sections or assigning tasks in shared projects. The repository itself calls the project an early stage, with more tools to come.

Who is the Todoist MCP Server for?

The server pays off for anyone who already uses Todoist and would rather capture and sort tasks by talking to an assistant. If you spend a lot of time in Claude Code or Codex, you no longer have to switch to the Todoist app. Teams can check project status without building reports. On cost: the server itself is free and open source. According to the help page, the integration works on every Todoist plan, including the free Beginner plan. In Claude Team or Enterprise, an Owner has to approve the connector first.

A Todoist account on any plan. Plus an MCP-capable client such as Claude, Claude Code, ChatGPT, Codex, Cursor or VS Code. For local use, Node.js and a Todoist API token. In Claude Team or Enterprise, an Owner must approve the connector.

The Todoist MCP Server gives your AI assistant access to your tasks in Todoist. It runs hosted at https://ai.todoist.net/mcp or locally on your machine. This guide covers both routes and what to keep in mind afterwards.

Step 1: Check first

You need a Todoist account. According to Todoist, the integration works on every plan, including the free Beginner plan. If you use Claude on a Team or Enterprise plan, an Owner has to approve the connector for the organization. Until then, the connection shows as Requested.

Claude in the browser and desktop app

Click the plus icon in the lower left and choose Connectors, then Add connector and Browse connectors. Search for Todoist and click Connect. Sign in to Todoist and grant access. Then open a new chat, because the tools load at the start of each conversation. Alternatively, add a custom connector with the server URL under Settings, Connectors.

Step 2: Claude Code

The quickest route is the official plugin. In Claude Code, run /plugin marketplace add doist/todoist-mcp and then /plugin install todoist@doist. The first time you use a Todoist tool, your browser opens to sign in. Without the plugin, add the server with claude mcp add --transport http todoist https://ai.todoist.net/mcp, launch claude, run /mcp and pick todoist to authenticate.

claude mcp add --transport http todoist https://ai.todoist.net/mcp

ChatGPT and Codex

Open ChatGPT in the browser or desktop app and choose Plugins in the sidebar. Search for Todoist, install the plugin and sign in. After that, Todoist is available in Codex as well. Call it there with @ or simply describe what you want in plain language. Codex asks before it writes anything.

Cursor and VS Code

In Cursor, add a todoist entry to ~/.cursor/mcp.json or to .cursor/mcp.json in your project. Set command to npx and args to -y, mcp-remote and https://ai.todoist.net/mcp. In VS Code, open the Command Palette, choose MCP: Add Server and then HTTP with the server URL.

Step 3: Local with an API token

To run the server yourself, you need Node.js and a Todoist API token from the settings under Integrations. Put the command npx @doist/todoist-mcp into your client's MCP configuration and set the TODOIST_API_KEY environment variable. Alternatively, npx -p @doist/todoist-mcp todoist-mcp-http starts a local HTTP server on port 3000. That helps if the hosted service keeps dropping the connection. Do not expose this server on a network without protection, because it runs with your token.

npx @doist/todoist-mcp
npx -p @doist/todoist-mcp todoist-mcp-http

Work safely

The server can change and delete tasks and projects. Review changes before you approve them. For recurring tasks, ask the assistant to reschedule rather than change the date, or the recurrence will be lost. You can see and remove granted access in Todoist under Settings, Integrations.

Hosted: Todoist OAuth in the browser, no token needed. Local: personal Todoist API token in the TODOIST_API_KEY environment variable; all tools then run as that Todoist user. You can review and revoke granted access in Todoist under Settings, Integrations.

Read and write access to the entire Todoist account: tasks, projects, sections, labels, filters, comments, reminders and assignments. The server can also delete objects. In workspaces, the signed-in user's permissions apply.

With the hosted service, the server runs at Todoist; locally, it runs on your machine. Task data the assistant reads goes to the respective AI provider. According to Todoist, its own privacy policy applies on the Todoist side and the assistant provider's policy on the other.

Medium, because the server can change and delete tasks and projects. A misunderstood request or a crafted task text with hidden instructions can trigger unwanted changes. Review write actions before approving them. Do not expose the local HTTP server on a network without your own access control, since it runs with your API token.