Todoist AI Agent Connector MCP Server

Overview

The Todoist AI Agent Connector MCP Server lets AI agents (LLMs and assistants) access and manage a Todoist account on behalf of a user. It provides a collection of reusable tools and a drop-in MCP (Model Context Protocol) server that exposes those tools over a streamable HTTP endpoint. The project is intended for developers who want to enable task management workflows from conversational agents, or to embed Todoist-aware tools into custom AI interfaces.

Tools are written to be reusable: they can be plugged into an MCP server, used directly inside your application, or wrapped for other AI SDKs. The server also includes compatibility helpers for OpenAI-style MCP clients and supports interactive MCP Apps (rich widgets) so responses can include structured UI components such as task lists.

Features

• Reusable Todoist tools (search, create, update tasks) that can be run locally or exposed via MCP
• Streamable HTTP MCP server endpoint for easy integration with modern AI clients
• OpenAI MCP-compatible search and fetch tools (JSON-encoded results)
• Support for MCP Apps (interactive UI widgets rendered in chat clients)
• Tool wrappers to inject a Todoist API client into each tool
• Local developer commands for testing, running single tools, and live development

Installation / Configuration

Install the package:

npm install @doist/todoist-ai

Run the built-in MCP server directly with npx (quick start):

npx @doist/todoist-ai

Environment

• Set TODOIST_API_KEY to a Todoist API token for local tool runs or to test the server.
• Use a test account or a temporary project when trying write operations.

Example: run a single tool locally (uses .env for TODOIST_API_KEY):

# Run a tool by name. Use -- to pass args to the script.
npm run tool -- findTasksByDate '{"query":"today"}'

Development / common scripts

npm start        # Build and run MCP inspector / server
npm run dev      # Development mode (auto rebuild + restart)
npm run tool:list # List available tools

Available Resources / Tools

The toolkit provides a set of tools in src/tools that implement common Todoist workflows. Example tools you’ll typically find:

• findTasksByDate — locate tasks based on date filters
• addTasks — create tasks from structured input
• search / fetch — OpenAI MCP-style tools returning JSON results
• user-info — check which Todoist account the API key is connected to

See the repository for a full, up-to-date list: https://github.com/doist/todoist-ai (browse src/tools)

MCP Apps

• The project supports rendering MCP Apps (inline widgets) for richer output (task lists, project summaries).
• See the docs/mcp-apps.md in the repo for details on widget structure and the build pipeline.

Integrations / Client Configuration

Primary MCP endpoint (streamable HTTP):

• https://ai.todoist.net/mcp

Quick integration snippets

Wrap tools for any AI SDK by injecting a Todoist client:

import { findTasksByDate, addTasks } from '@doist/todoist-ai'
import { TodoistApi } from '@doist/todoist-sdk'

const todoist = new TodoistApi(process.env.TODOIST_API_KEY)

function wrapTool(tool) {
  return {
    ...tool,
    execute(args) { return tool.execute(args, todoist) },
  }
}

Client-specific configuration examples (quick reference)

Example Cursor mcp.json:

{
  "mcpServers": {
    "todoist": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://ai.todoist.net/mcp"]
    }
  }
}

VS Code server snippet:

{
  "servers": {
    "todoist": {
      "type": "http",
      "url": "https://ai.todoist.net/mcp"
    }
  }
}

Use Cases

• Personal assistant automation: Let a conversational agent add, reschedule, or mark tasks complete based on natural-language instructions like “move all overdue tasks for Project X to tomorrow.”
• Meeting follow-ups: After a meeting summary, auto-create action items as Todoist tasks with due dates and assignees.
• Inbox triage: Use an agent to parse an email or notes and create prioritized tasks with labels and estimated due dates.
• Developer tooling: Integrate Todoist tools into a code editor assistant (VS Code/Claude/other MCP clients) so a developer can create tasks without switching context.

Concrete example — Vercel AI SDK

import { findTasksByDate, addTasks } from '@doist/todoist-ai'
import { TodoistApi } from '@doist/todoist-sdk'
import { streamText } from 'ai'

const client = new TodoistApi(process.env.TODOIST_API_KEY)
const tools = {
  findTasksByDate: wrapTool(findTasksByDate, client),
  addTasks: wrapTool(addTasks, client),
}

await streamText({ model: yourModel, system: 'You are a Todoist assistant', tools })

Safety & Development Notes

• When running write operations, prefer a test Todoist account or a disposable project.
• The toolkit is designed to be extensible — tools are plain modules you can import and adapt into your own AI workflows.
• For contributor and development practices, see the repo’s CONTRIBUTING.md and docs directory.

For full documentation, advanced setup (OAuth, custom MCP servers), and the latest tool list, consult the GitHub repository: https://github.com/doist/todoist-ai.