← Also built

MCP server

Tomato MCP

A small MCP server that lets an AI app such as Claude read and update my Tomato Timer task list.

  1. Step 01

    AI app

    Tested with Claude Desktop and Claude Code.

  2. Step 02

    Tomato MCP

    A small MCP server with four tools: get_context, list_tasks, add_task and update_task.

  3. Step 03

    GitHub API

    Reads and writes one task file in a private repository.

  4. Step 04

    Tomato Timer

    Changes show up on the next sync, and the other way round.

MCP is a standard way for an AI app to use outside tools. An AI app can get a short snapshot of today's tasks, list tasks, add a task to Today or Later, and mark one done, move it, star it or rename it.

It is a Python server (3.10 or newer, MIT licence). I built it with Claude Code. I decided what it should do, reviewed the work and tested it. I did not write it from scratch.

Guardrails

  • It never deletes tasks, and only changes the rows it edits.
  • It uses a token limited to that one repository. Read and write use separate scoped tokens.
  • If the app saves at the same moment, it reloads and tries again instead of overwriting, up to three times.

Honest outcome

Three automated tests that run without a network connection or a real token. It runs on the computer where it is installed. A hosted version for web-based AI apps does not exist yet.

How it connects

  • Diagram in four boxes joined by arrows. An AI app, tested with Claude Desktop and Claude Code, connects over MCP to Tomato MCP, a small MCP server with four tools that never deletes tasks. That connects through the GitHub API to one task file in a private GitHub repository, which syncs with my Tomato Timer app.

    How Tomato MCP connects an AI app to my task list. Drawn from the project's public README.

View on GitHub