Official Todoist AI tool library and MCP Server providing 40+ tools for LLM agents to fully control Todoist accounts.
@doist/todoist-ai is an official Doist-maintained TypeScript library designed to expose the full range of Todoist task management capabilities to AI agents. The project serves as a standard MCP (Model Context Protocol) Server while keeping tool definitions decoupled from the transport layer, enabling integration as a standalone function library into any AI framework.
The toolkit includes 40+ tools covering task CRUD and completion/uncompletion, project management, labels/filters/sections, comments and reminders, productivity analysis and project health insights, search and attachment retrieval, and team collaboration with task assignment. The design emphasizes workflow-level tools (e.g., get-overview, analyze-project-health) to balance flexibility with LLM call efficiency.
It provides a dual-mode MCP Server: a local stdio mode launchable via npx @doist/todoist-ai, and a remote Streamable HTTP mode served at the official hosted endpoint https://ai.todoist.net/mcp with OAuth support. Verified compatibility includes Claude Desktop, Cursor, Claude Code CLI, and VS Code. Additionally, MCP Apps — interactive UI widgets built on @modelcontextprotocol/ext-apps and React 19 — enable inline rich component rendering within AI chat interfaces.
Under the hood, it depends on @doist/todoist-sdk for Todoist API calls, Zod v4 for parameter validation, Vite for building, Vitest for testing, and semantic-release for automated publishing. Licensed under MIT, currently at v8.12.1.