🏃 Garmin MCP Server

OpenClaw Platform · Port 8422

FastMCP server for Garmin Connect — live GPS tracking, health metrics, and activity data for AI agents.

⚡ Quick Start

# 1. Clone / enter project
cd garmin-mcp

# 2. Install dependencies
pip install -r requirements.txt

# 3. Configure credentials
cp .env.example .env
# Edit .env → set GARMIN_EMAIL and GARMIN_PASSWORD

# 4. Run server
python server.py

Server starts at: http://localhost:8422

First run logs into Garmin Connect and caches the session token at ~/.garmin_mcp_tokens. Subsequent runs reuse the cached token (no re-login needed).

Related MCP server: GC-MCP

🛠️ Tools (12 total)

📡 LiveTrack Real-time GPS

1. On your Garmin device → start activity → enable LiveTrack

2. Garmin sends a share URL to your contacts, e.g.: https://livetrack.garmin.com/session/abc123.../token/xyz...

3. Extract the session token (between /session/ and /token/)

4. Call garmin_get_livetrack with that token — positions update ~every 4 seconds

🔌 OpenClaw Integration

Add to your OpenClaw config:

skills:
  - name: garmin-health
    transport: streamable-http
    url: http://localhost:8422

Or connect via MCP protocol URL in Claude Desktop:

http://localhost:8422/sse

🔐 Auth Notes

• Uses the unofficial garminconnect Python library (no Garmin Developer Program approval needed)

• Tokens cached at ~/.garmin_mcp_tokens — delete this file to force re-login

• Garmin may prompt for 2FA on first login in some regions

• Rate limiting: avoid calling more than ~100 times/hour to prevent IP blocks

📁 Project Structure

garmin-mcp/
├── server.py          # FastMCP server (12 tools)
├── garmin_client.py   # Auth + session wrapper
├── models.py          # Pydantic v2 input models
├── utils.py           # Markdown/JSON formatters
├── requirements.txt
├── .env.example
└── README.md

🌐 Response Formats

All tools support response_format:

• "markdown" (default) — human-readable tables, great for chat UIs

• "json" — raw Garmin API data, great for downstream processing

Built for the Dr. Arty+ Team · OpenClaw Platform · 2026