In progress

Denyut

A health assistant for Amazfit watches. An MCP server with 26 tools lets Claude or ChatGPT answer questions from your own data.

Zepp OS mini program, JavaScript, Bun, TypeScript, SQLite, MCP, Caddy

Visit site
  • Case study summary: problem, role, and result

  • How it works: from the watch to any AI assistant, with key features

  • Denyut website home page with a sample chat

  • Consent screen on the watch

  • Sync screen on the watch

  • Watch screen suggesting a question for the AI assistant

1 / 6

The problem

Smartwatches record a lot of health data, but the numbers and terms can be hard to understand. Denyut lets people ask about that data in plain language.

My role

I designed and built the watch app, the server, the MCP connector, and the website.

How it works

A Zepp OS app on the watch sends data through the Zepp phone app to the Denyut server. The server stores it in SQLite and exposes it through MCP.

Users pair the watch with an 8-digit code and sign in with Google, then connect the AI assistant they prefer, such as Claude, ChatGPT, or Gemini.

Key features

  • MCP server with 26 tools for sleep, heart rate, stress, SpO2, activity, and trends
  • Comparison with each user's own baseline and early warning signs
  • Accuracy notes that flag doubtful data without changing the numbers
  • Available on the Zepp App Store

Denyut is not a medical device and is not an official Zepp or Amazfit product. The chat shown here is a sample.