# Todoist MCP Server [Official] [Health: Active]

**Category:** 🏢 Workplace & Productivity  
**Repository:** https://github.com/Doist/todoist-mcp  
**GitHub Stars:** 563  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/todoist-mcp-server

## Description
Todoist's official MCP server from Doist. Lets AI agents find, add, update and complete your Todoist tasks and projects. Use the hosted endpoint with OAuth, or run @doist/todoist-mcp locally.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "todoist-mcp-server": {
    "command": "npx",
    "args": ["-y","@doist/todoist-mcp"],
    "env": {
      "TODOIST_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `TODOIST_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Todoist MCP Server does

Todoist MCP Server gives AI agents access to Todoist accounts through the Model Context Protocol. Its task and project workflows cover finding tasks, adding new tasks, updating existing items, and marking tasks complete. The server is maintained by Doist and uses the official Todoist TypeScript API client.

The package is also usable as a library. Applications can import individual tools and attach them to their own AI interface instead of communicating with the MCP server. The repository demonstrates this pattern with `findTasksByDate` and `addTasks` used through Vercel's AI SDK.

## How it works

The hosted deployment exposes a streamable HTTP MCP endpoint at `https://ai.todoist.net/mcp`. Clients that support HTTP MCP connections can connect there and authenticate with Todoist through OAuth. Todoist access can be revoked later from Todoist's Settings under Integrations.

For local use, launch the package with `npx @doist/todoist-mcp`. The repository also supports direct tool execution for development and testing. In that mode, the Todoist API client reads `TODOIST_API_KEY`, and the project recommends checking the connected account with the `user-info` tool before performing write operations.

The server includes `search` and `fetch` tools that follow OpenAI's MCP specification. Their results are JSON-encoded for OpenAI-compatible use while remaining usable within the broader MCP ecosystem.

## Setup and configuration

For the hosted service, configure an MCP client with the streamable HTTP URL and complete the OAuth flow. Claude Desktop uses its custom connector flow; Cursor can use `mcp-remote`; Claude Code supports either the official Todoist plugin or a manual HTTP MCP entry; Gemini CLI provides an official Todoist extension; and Visual Studio Code can add the URL as an HTTP server.

To use the package locally, install or invoke it with npm. For repository development, `npm run setup` creates a `.env` file from `.env.example`; direct tool execution then uses the API key in that file. Test accounts or temporary projects are appropriate for write-operation checks because the tools can modify real Todoist data.

## Tools and capabilities

- Find Todoist tasks, including date-based lookup through `findTasksByDate`.
- Add tasks with `addTasks`.
- Update and complete Todoist tasks and projects.
- Expose `search` and `fetch` for OpenAI MCP compatibility.
- List available tools with `npm run tool:list`.
- Execute a selected tool with JSON arguments through `npm run tool -- <tool-name> '<json-args>'`.
- Render interactive MCP App widgets for tool output, including task-list views.

The repository's tools are not tied exclusively to MCP. Developers can import them into other conversational AI applications and provide a Todoist API client when executing them.

## Limitations and notes

The project describes itself as being in an early stage, with a small tool set intended to support complete workflows. The hosted OAuth path does not require a local API key, while local direct execution uses `TODOIST_API_KEY`. Write operations change Todoist data, so account and project selection should be checked before testing them. The MCP Apps support depends on AI clients that can render inline interactive widgets; clients may otherwise receive ordinary tool results.

_Full upstream README: https://allmcps.com/mcp/todoist-mcp-server/readme_