Garmin MCP

A Model Context Protocol server that exposes your Garmin Connect data (steps, sleep, HR, HRV, stress, body battery, activities, training readiness, body composition) to MCP-compatible clients like Claude.

Built with FastMCP and python-garminconnect. Runs over HTTP with bearer-token auth — ready to deploy to Render.

Tools

| Tool | What it returns | | --- | --- | | get_user_profile | Display name, full name, unit system | | get_daily_stats | Steps, calories, resting HR, stress, body battery, intensity minutes | | get_steps_intraday | Per-bucket step samples for one day | | get_daily_steps | Daily totals across a date range | | get_heart_rates | HR samples + resting HR | | get_sleep_data | Stages, scores, SpO2, HRV, respiration | | get_stress_data | Stress levels, rest vs. active durations | | get_body_battery | Body battery values | | get_body_composition | Weight, BMI, body fat, muscle mass | | get_floors | Floors ascended / descended | | get_hrv_data | Overnight HRV | | get_training_readiness | Training readiness scores | | get_recent_activities | Most recent activities | | get_activities_by_date | Activities in a date range, optional type filter | | get_last_n_days_summary | Daily stats for the last N days |

Local dev

pip install -r requirements.txt
cp .env.example .env
# fill in GARMIN_EMAIL, GARMIN_PASSWORD, MCP_BEARER_TOKEN
set -a && source .env && set +a
python server.py

Hit http://localhost:8000/mcp to confirm the server is responding (you'll get a 4xx from a plain GET — that's expected; MCP clients do POST/SSE).

Deploy to Render

One click deploys the render.yaml blueprint — Render then prompts you for the environment variables below. Or do it manually:

1. Push this repo to GitHub.
2. On render.com: New → Web Service → Connect GitHub → pick this repo. Render auto-detects render.yaml.
3. In the Environment tab, fill in:
   • GARMIN_EMAIL
   • GARMIN_PASSWORD
   • GARMINTOKENS_BASE64 (only if your account has MFA — see below)
   • MCP_BEARER_TOKEN is auto-generated; copy it from the dashboard.
4. Deploy. When the build is green, your URL is https://<service>.onrender.com/mcp.

Handling MFA

The server runs non-interactively, so it can't prompt for an MFA code. Bootstrap the tokens once locally:

pip install garminconnect
python bootstrap_tokens.py
# enter email + password + MFA code when asked
# copy the printed GARMINTOKENS_BASE64 blob into Render

The Garmin OAuth refresh token is good for ~1 year — set it once and forget.

Connect to Claude

Desktop / web (claude.ai) → Settings → Connectors → Add custom connector:

• URL: https://<service>.onrender.com/mcp
• Name: Garmin
• Authentication: Bearer token → paste MCP_BEARER_TOKEN

Toggle the connector on inside a chat to use the tools.

Security notes

• MCP_BEARER_TOKEN is the only thing between the internet and your Garmin data — generate a long random value and rotate if leaked.
• Garmin credentials are kept in env vars, never written to the repo.
• Token cache lives under /tmp/.garminconnect on the server; Render's filesystem is ephemeral, so the server re-uses cached tokens within an instance lifetime and re-bootstraps from GARMINTOKENS_BASE64 on cold starts.