Project 4: Ask an AI about your training

Goal: an AI assistant such as Claude answering questions about your sleep, heart and training from your real numbers, through a token that can only read.

Time: 10 minutes. Level: copy and paste. Needs: project 1, and an AI client that speaks MCP (Claude Desktop, Claude Code, or any other MCP client).

1. Get a read-only token and the lines for your clientCopy Anchor Link

python3 recipes/ai-agent.py

Read only token, valid until 2026-11-11. Keep it private.

Claude Code:

  claude mcp add --transport http open-health-sync http://localhost:8080/mcp \
    --header "Authorization: Bearer ohs_ut_..."

Claude Desktop, or any client that starts a local command (claude_desktop_config.json):

{ "mcpServers": { "open-health-sync": { "command": "ohs", "args": ["mcp"], ... } } }

Paste the lines for your client and restart it. Claude Desktop starts the ohs command you installed in project 1, so it needs to be on your PATH.

The token reads your data and nothing else: it cannot write, delete or connect an account, and it stops working after 30 days. Before you connect it, know where your data goes: what the assistant reads to answer you is sent to the company that runs the model, like anything you type to it. Your instance itself stays on your machine.

2. Ask your first questionsCopy Anchor Link

Start broad, so the assistant finds out what you have:

What health data can you see for me? List each source, each kind of data and the dates it covers.

Then ask a real question:

Compare my sleep on the nights after I trained with the nights after rest days, over the last eight weeks. Show the averages and how many nights each one is based on.

Has my resting heart rate changed since I started running more in September? Show it week by week.

The assistant calls the server's tools (get_sleep, get_workouts, get_daily, get_samples and the others) and works from what they return. Ask it to show its numbers: you can check any of them with the scripts of projects 2 and 3.

3. Keep the good questionsCopy Anchor Link

When a question gives you something useful, save it. The prompt library has a dozen to start from, written for people who train: a morning check, a weekly review, race week, coming back from a break.

Make it yoursCopy Anchor Link

• Ask for a chart: most assistants can turn the numbers into a picture or an HTML page.
• Ask it to write a recipe: "write a Python script like recipes/sleep-report.py that shows my workouts per week". Then run it and read it.
• Shorten the token's life: python3 recipes/ai-agent.py --days 1 for a one-off session.

Share itCopy Anchor Link

The best question you asked, and what made the answer useful, in Show and tell. Good prompts make it into the library.

Next: Project 5, Does training cost you sleep?.