Apple Health in Claude Code

What you need

• An iPhone on iOS 15 or later with Apple Health. Apple Watch data comes along too.
• The HealthAPI app. Free for 2 days, no card needed. Then an optional 7-day free trial when you subscribe. The preview includes 25 requests a day. When the 2 days are up, AI apps and the API stop reading your data until you choose a plan in the app. Your synced data is kept for about 5 weeks, so subscribing picks up where you left off. Pro is $4.99 a month or $49.99 a year (100 requests a day); Developer is $9.99 a month or $99.99 a year (1,000 a day). Cancel anytime in the App Store.
• Claude Code on your computer.

Set it up in 4 steps

1. 1

   Install HealthAPI and sync

   Get HealthAPI from the App Store, open it and allow Apple Health. Your history uploads, then new data keeps syncing.
2. 2

   Add the server

   In a terminal, run claude mcp add --transport http --scope user healthapi https://healthapi.app/mcp. The user scope makes it available in every project.
3. 3

   Sign in and approve on your iPhone

   Start claude, run /mcp, choose healthapi and sign in. A HealthAPI page opens in your browser with a short code. Scan its QR code with your iPhone, or enter the code in HealthAPI under Connect, check it matches, and tap Approve.
4. 4

   Ask Claude Code

   Ask for something that needs your data, such as “Use HealthAPI to get my resting heart rate for the last 60 days and chart it.” Claude Code asks before each tool call unless you allow them.

claude mcp add --transport http --scope user healthapi https://healthapi.app/mcp

The sign-in page names Claude Code with a check mark and notes that it returns to a program on this computer: Claude Code waits for the answer on localhost. If you started the sign-in yourself and the codes match, approve it. From a shell, claude mcp login healthapi does the same as /mcp.

Scopes, projects and API keys

--scope user keeps the server in ~/.claude.json for all your projects. Leave it out and it applies to the current project only. To share it with a repository, use the project scope, which writes .mcp.json:

claude mcp add --transport http --scope project healthapi https://healthapi.app/mcp

That file holds only the address, so it's safe to commit. Everyone who opens the project signs in with their own iPhone and sees only their own data.

Rather use your API key than sign in? Copy it in the app (Connect, then Other tools), put it in an environment variable, and reference the variable. Claude Code fills in ${HEALTHAPI_KEY} when it connects, so the key itself never lands in the file:

{
  "mcpServers": {
    "healthapi": {
      "type": "http",
      "url": "https://healthapi.app/mcp",
      "headers": { "Authorization": "Bearer ${HEALTHAPI_KEY}" }
    }
  }
}

Fewer permission prompts

Claude Code asks before every HealthAPI tool call. The tools only read, so you can allow them all in ~/.claude/settings.json, or for one tool at a time with its full name:

{
  "permissions": {
    "allow": ["mcp__healthapi__*"]
  }
}

Each call still uses one of your daily requests, so an agent working unattended can spend them. Raw readings from get_samples can also be long; asking for summaries keeps the output small.

What to build with it

• A dashboard in your repo. “Use HealthAPI to get 90 days of sleep, resting heart rate and HRV, save it to data/health.json, and build a static page that charts it.”
• A script that runs without you. “Look at one HealthAPI daily summary, then write a Node script that reads yesterday from the REST API with HEALTHAPI_KEY and appends it to a CSV.”
• An analysis you can check. “Compare my sleep after workout days with rest days over the last 3 months, and show the numbers you used.”
• Fixtures for your own app. “Fetch my last weekly summary and turn it into TypeScript types and a test fixture.”

Once you've signed in, non-interactive runs can use the tools too. Allow just the ones the prompt needs:

claude -p "Use HealthAPI to summarize last night's sleep in three lines" \
  --allowedTools "mcp__healthapi__get_sleep"

Code that runs on a schedule doesn't need Claude at all: call the REST API with your key. The JavaScript and Python guides have starting points.

The tools, by their Claude Code names

Every tool is read only. The MCP server page covers inputs, sign-in and discovery, and the Cursor and Codex guides cover the same server in other tools.

Questions