Google Health MCP

Ask your AI assistant about your own health data — steps, sleep, runs, heart rate, weight and more. Self-host in ~10 minutes on Cloudflare's free tier.

Works with Claude, Cursor, Windsurf, Zed, VS Code (Copilot), Continue.dev, and any other MCP-compatible client.

See it in action

Monthly exercise recap

You:    Summarise my exercise sessions for May

AI:     You logged 39 sessions in May — 2 runs and 37 walks.

        Highlights:
        • May 9  — Long run: 21.4 km in 2h 56m (looks like a half marathon effort!)
        • May 5  — Run: 5.1 km in 39 min
        • May 25 — Most active day: 7 walks, ~6.5 km total

        You were most active in the last week of May (May 22–31),
        with activity on 8 out of 10 days.

Sleep check-in

You:    How did I sleep last week?

AI:     Here's your sleep for Jun 16–22:

        Mon  7h 42m  ██████████████████░░░░
        Tue  6h 55m  ████████████████░░░░░░
        Wed  8h 03m  ████████████████████░░
        Thu  7h 20m  ██████████████████░░░░
        Fri  6h 30m  ███████████████░░░░░░░
        Sat  8h 45m  █████████████████████░
        Sun  7h 55m  ████████████████████░░

        Average: 7h 36m. Saturday was your best night.
        Tuesday and Friday were slightly under 7 hours.

Progress towards a goal

You:    Am I hitting 10,000 steps a day this month?

AI:     Over the last 30 days your daily average is 8,340 steps —
        about 83% of your 10,000-step target.

        You hit the goal on 11 out of 30 days.
        Best day: June 20 with 14,200 steps.
        Weekends tend to be stronger than weekdays for you.

Heart rate during a run

You:    What was my heart rate during my run on June 24?

AI:     During your 27-minute run on June 24 (3.76 km):

        Average heart rate: 180 bpm
        That puts it firmly in Zone 4 (threshold/hard effort).
        Your resting HR that day was 58 bpm — roughly 3× baseline.

Related MCP server: Google Health MCP

More things you can ask

• "How has my resting heart rate changed over the last 3 months?"

• "Did I average more than 7 hours of sleep in June?"

• "Compare my step counts this week vs last week."

• "What's my average SpO2 at night?"

• "How much have I walked vs run this year?"

• "Show me my weight trend since January."

• "Which day of the week am I most active?"

Available tools

Activity & Fitness

Cardio & Heart

Health Metrics

Sleep

Nutrition

Utilities

Prerequisites

• Cloudflare account (free tier is sufficient)

• Google Cloud project with the Health API enabled

• Node.js 18+

Setup

7 steps, ~10 minutes.

1. Clone and install

git clone https://github.com/akshaygoyal/google-health-mcp.git
cd google-health-mcp
npm install

2. Enable the Google Health API

1. Go to Google Cloud Console → APIs & Services → Enable APIs

2. Search for and enable Google Health API

3. Go to Credentials → Create OAuth 2.0 Client ID (type: Web application)

4. Add http://127.0.0.1:8765/callback as an authorised redirect URI

5. Note your Client ID and Client Secret

3. Configure wrangler.toml

cp wrangler.toml.example wrangler.toml
npm run kv:create

Copy the returned id and paste it into wrangler.toml under [[kv_namespaces]]. This file is gitignored so your KV namespace ID stays off GitHub.

4. Set secrets in Cloudflare

# A long random string — becomes part of your private connector URL
wrangler secret put MCP_SHARED_SECRET

# From your Google Cloud OAuth client
wrangler secret put GOOGLE_CLIENT_ID
wrangler secret put GOOGLE_CLIENT_SECRET

5. Get your Google refresh token

Choose a short ID for yourself (letters, numbers, hyphens, underscores — e.g. alice or me):

GOOGLE_CLIENT_ID=your-id GOOGLE_CLIENT_SECRET=your-secret npm run token:setup -- --user=alice

This opens a browser for Google consent. After approving, copy the printed command and run it:

wrangler kv key put --binding=HEALTH_TOKENS "user:alice:refresh_token" "your-refresh-token"

6. Deploy

npm run deploy

Your MCP server is now live. Each user's connector URL is:

https://google-health-mcp.<your-workers-subdomain>.workers.dev/mcp/<userId>/<MCP_SHARED_SECRET>

For example, the user alice would connect to:

https://google-health-mcp.<your-workers-subdomain>.workers.dev/mcp/alice/<MCP_SHARED_SECRET>

7. Connect your MCP client

Add your personal URL as a custom MCP server in your AI client:

• Claude.ai → Customize → Integrations → Add integration URL

• Cursor → Settings → MCP → Add server URL

• Windsurf → Settings → MCP Servers → Add

• VS Code (Copilot) → .vscode/mcp.json → add server entry

• Continue.dev → config.json → mcpServers array

Adding more users

Each additional user needs their own Google consent and their own KV entry — the Worker and all secrets are shared.

# On the new user's machine (they need your GOOGLE_CLIENT_ID / GOOGLE_CLIENT_SECRET):
GOOGLE_CLIENT_ID=your-id GOOGLE_CLIENT_SECRET=your-secret npm run token:setup -- --user=bob

Run the printed wrangler kv key put command to store their refresh token:

wrangler kv key put --binding=HEALTH_TOKENS "user:bob:refresh_token" "bobs-refresh-token"

Bob's connector URL is then:

https://google-health-mcp.<your-workers-subdomain>.workers.dev/mcp/bob/<MCP_SHARED_SECRET>

No redeployment needed — the new KV entry is picked up immediately.

Security note: MCP_SHARED_SECRET is a shared gate — anyone who has any user's URL already knows the secret and could construct URLs for other user IDs. Only share the server with people you trust, and keep the secret itself private.

Local development

cp wrangler.toml.example wrangler.toml  # if not done already
cp .dev.vars.example .dev.vars
# Fill in both files with your credentials
npm run dev

Data types

All 38 Google Health API data types have dedicated tools. The get_raw_data_points tool is also available for querying any type by its raw identifier — see the Google Health data types reference for the full list.

Security

• Each user's MCP endpoint URL (/mcp/<userId>/<MCP_SHARED_SECRET>) is their only credential — treat it like a password

• MCP_SHARED_SECRET is a shared worker-level secret; userId is a namespace, not an additional auth factor — see the note in Adding more users

• This server is read-only: it never writes data back to Google Health

• Refresh tokens are stored in Cloudflare KV, encrypted at rest, namespaced per user

Staying up to date

To pull the latest changes and redeploy your instance:

git pull origin main
npm install   # only needed if dependencies changed
npm run deploy

No re-setup of secrets or KV is needed — those persist across deployments.

If new tools aren't showing up in your AI client after a deployment, disconnect and reconnect the integration — most MCP clients cache the tool list from when you first connected.

If a release changes the Google OAuth scopes (check the CHANGELOG), each user will need to re-run npm run token:setup -- --user=<userId> to get a new refresh token with the updated permissions.

Changelog

See CHANGELOG.md for a full history of releases and what changed in each version.

Feedback & contributions

Tried it? Found a bug? Want a new data type added?

👉 Open an issue — feedback of any kind is very welcome, especially from first-time users.

If you'd like to contribute, please read the contributing guide first.

License

MIT