A Model Context Protocol (MCP) server that gives compatible AI clients access to Garmin Connect activity and health data stored in SQLite, plus tools for managing Garmin workouts.
npm installCreate a .env file in the project root:
cp .env.example .envEdit .env and add your Garmin Connect credentials:
GARMIN_USERNAME=your-email@example.com
GARMIN_PASSWORD=your-password
Create the local data directory and download your activities:
mkdir -p data
npm run downloadThis downloads your activities and lap data from Garmin Connect to a local SQLite database at data/garmin-data.db.
First, get the absolute path to this project:
pwdThis will output something like /Users/yourusername/Code/garmin-mcp-server.
macOS: Edit ~/Library/Application Support/Claude/claude_desktop_config.json
Linux: Edit ~/.config/Claude/claude_desktop_config.json
Windows: Run cd to get the path, then edit %APPDATA%\Claude\claude_desktop_config.json
Add this configuration (replace the path with your actual project path):
{
"mcpServers": {
"garmin-mcp-server": {
"command": "npx",
"args": ["-y", "tsx", "/Users/yourusername/Code/garmin-mcp-server/src/index.ts"],
"cwd": "/Users/yourusername/Code/garmin-mcp-server"
}
}
}Replace /Users/yourusername/Code/garmin-mcp-server with your actual path from the pwd command.
After restarting, you can ask Claude:
- "Sync my latest Garmin activities"
- "Sync my daily resting heart rate"
- "Show me my 5 most recent runs"
- "What's my average pace this month?"
- "Analyze my running performance trends"
- "How many kilometers did I run this year?"
- Local SQLite database - Activity, lap, and daily heart-rate data for fast access
- Incremental sync - Update Garmin data through an MCP client or the command line
- Read-only SQL tools - Analyze training data with custom
SELECTqueries - Workout management - Create, rename, schedule, unschedule, and delete Garmin workouts
- Two transports - Local stdio and self-hosted Streamable HTTP
- Optional HTTP authentication - Protect remote access with a bearer token
Deploy your own instance using Docker for use with Claude, ChatGPT, or any MCP-compatible client.
- Clone this repository:
git clone https://github.com/vnglst/garmin-mcp-server.git
cd garmin-mcp-server- Create a
.envfile and replace the example values:
cat > .env <<'EOF'
GARMIN_USERNAME=your-email@example.com
GARMIN_PASSWORD=your-password
API_KEY=your-secret-api-key
EOF- Create the initial database on the host. The production image does not include the download CLI, so the database must exist before the container starts:
npm ci
mkdir -p data
npm run download- Create the container, copy the database into its persistent volume, and start it:
docker compose create
docker compose cp ./data/garmin-data.db garmin-mcp-server:/app/data/garmin-data.db
docker compose up -dThe Compose configuration builds the image from the checked-out source. The server runs at http://localhost:3000 with:
- MCP endpoint:
POST /mcp - Health check:
GET /health
- Create a new service from Git repository
- Set the following environment variables:
GARMIN_USERNAME: Your Garmin Connect emailGARMIN_PASSWORD: Your Garmin Connect passwordAPI_KEY: Secret key for authenticationPORT: 3000 (default)
- Configure health check path:
/health - Mount persistent storage at
/app/dataand seed it withgarmin-data.db - Deploy
Initialize a named volume from a database created with npm run download before starting the image:
docker volume create garmin-data
docker run --rm \
-v "$PWD/data:/source:ro" \
-v garmin-data:/target \
alpine cp /source/garmin-data.db /target/garmin-data.dbdocker run -d \
--name garmin-mcp-server \
-p 3000:3000 \
-e GARMIN_USERNAME=your-email@example.com \
-e GARMIN_PASSWORD=your-password \
-e API_KEY=your-secret-key \
-v garmin-data:/app/data \
ghcr.io/vnglst/garmin-mcp-server:latestConfigure your MCP client to connect to the HTTP endpoint:
{
"mcpServers": {
"garmin": {
"url": "https://your-server.example.com/mcp",
"transport": "http",
"headers": {
"Authorization": "Bearer your-api-key"
}
}
}
}| Variable | Required | Description |
|---|---|---|
GARMIN_USERNAME |
For sync/workout tools | Garmin Connect email |
GARMIN_PASSWORD |
For sync/workout tools | Garmin Connect password |
API_KEY |
No | Bearer token required by HTTP clients; strongly recommended outside an isolated local environment |
PORT |
No | Enables HTTP mode on this port; defaults to 3000 in Docker |
HTTP_MODE |
No | Set to true to enable HTTP mode without setting PORT |
CORS_ORIGIN |
No | Comma-separated allowed browser origins; CORS headers are omitted by default |
MAX_SESSIONS |
No | Maximum active HTTP MCP sessions (default: 100) |
MAX_QUERY_CHARS |
No | Maximum SQL query length (default: 50000) |
The server exposes activity-data tools and Garmin Connect workout-management tools.
Get the database schema to understand available data fields.
Example usage in Claude:
- "What data fields are available in my running database?"
- "Show me the database schema"
Execute SELECT queries against your activities database.
Example usage in Claude:
- "Show me my 10 most recent runs"
- "What's my average heart rate this month?"
- "Find all runs longer than 10km"
- "Calculate my total distance this year"
Security: One SELECT or WITH ... SELECT statement is allowed per call. Write statements and SQLite administrative statements are blocked.
Download and sync new activities from Garmin Connect to the local database.
Example usage in Claude:
- "Sync my latest Garmin activities"
- "Update my running data"
- "Check for new activities"
Returns a summary showing:
- Number of new activities downloaded
- Total activities in the database
- Date of your latest activity
Incrementally sync daily heart-rate summaries from Garmin Connect. The first sync
downloads the latest 30 days; later syncs refresh the newest stored day and add
new days through today. Use startDate and endDate in YYYY-MM-DD format to
backfill a specific range.
The daily_heart_rate table includes resting, minimum, maximum, and seven-day
average resting heart rate. Metrics unavailable from a device are stored as
NULL and can be queried with run-query.
The following tools manage structured workouts in Garmin Connect:
| Tool | Description |
|---|---|
list-workouts |
List saved workouts |
get-workout |
Get a workout and all of its structured steps |
create-workout |
Create a workout from a Garmin workout JSON object |
rename-workout |
Rename a workout without changing its steps |
delete-workout |
Permanently delete a saved workout |
schedule-workout |
Add a workout to the Garmin calendar |
list-scheduled-workouts |
List calendar entries for a month |
get-scheduled-workout |
Get a calendar workout by schedule ID |
unschedule-workout |
Remove a calendar entry without deleting the workout |
Examples:
- "Create a 5 x 3-minute running interval workout and schedule it tomorrow"
- "Rename my VO2 max workout"
- "Remove the duplicate scheduled workout but keep the saved workout"
Workout creation, renaming, scheduling, unscheduling, and deletion modify your
Garmin Connect account. delete-workout removes the saved workout itself;
unschedule-workout removes only one calendar entry.
The SQLite database contains three tables as data is synced:
activitiesfor activity summaries and performance metricsactivity_lapsfor per-lap pace, heart rate, cadence, running dynamics, elevation, and location metricsdaily_heart_ratefor daily minimum, maximum, resting, and seven-day-average resting heart rate
Use the get-schema tool to inspect the exact current columns. Activity data includes:
Basic Metrics:
- Activity ID, name, description, timestamps (local & GMT)
- Activity type, location name
- Distance, duration, elapsed duration, moving duration
- Calories, steps, lap count
Heart Rate Data:
- Average/max heart rate
- Lactate threshold BPM
- Time in each HR zone (1-5)
- VO2 Max value
Speed & Pace:
- Average/max speed
- Fastest splits (1K, 5K, 10K, mile)
Running Dynamics:
- Average/max stride length
- Average/max cadence (including double cadence)
- Average vertical oscillation
- Average ground contact time
- Vertical ratio, vertical speed
Training Load & Intensity:
- Activity training load
- Training effect (aerobic/anaerobic)
- Vigorous/moderate intensity minutes
Elevation:
- Elevation gain/loss
- Min/max elevation
Power Metrics:
- Average/max power
- Grit and Flow scores
- Initial Setup: Run
npm run downloadto download all your activities from Garmin Connect into a local SQLite database - MCP integration: The server exposes tools that compatible clients can use to query your data
- Stay updated: Use the
sync-activitiesandsync-healthtools to download new data anytime - Flexible queries: The client can run read-only SQL queries to analyze your data
# Download/update activities from Garmin
npm run download
# Build TypeScript to JavaScript
npm run build
# Run the unit test suite
npm test
# Run the stdio and HTTP smoke test (requires a build and local database)
npm run test:mcp
# Run server directly (for testing)
npm startNote: When using with Claude Desktop via the tsx configuration, you don't need to build - changes to TypeScript files are automatically picked up.
- Verify credentials in
.envfile (GARMIN_USERNAME and GARMIN_PASSWORD) - Ensure you can log into connect.garmin.com manually
- Try logging in via web browser first
- Check that your password doesn't contain special characters that need escaping
- Run
npm run downloadfirst to create the database - Check that
data/garmin-data.dbexists in your project directory - If the database may be corrupted, back it up before recreating it with
npm run download
- Restart Claude Desktop after configuration changes
- Check Claude Desktop logs for error messages
- Verify the paths in
claude_desktop_config.jsonare absolute, not relative - Ensure
.envfile exists in the project root directory
- Only SELECT queries are allowed (INSERT, UPDATE, DELETE are blocked)
- Use the
get-schematool to see available table columns - Check SQL syntax is correct
For a local stdio installation:
- Credentials are loaded from the local
.envfile and sent to Garmin Connect for authentication. - Synced data is stored in the local SQLite database.
- Query access is read-only.
For a self-hosted HTTP installation, credentials and data reside on the host running the container, and MCP responses travel between that host and the configured client. Set a strong API_KEY, use HTTPS at the reverse proxy, restrict network access, and protect the persistent volume and .env file. Workout tools can modify the connected Garmin account; review destructive tool calls before approving them.
Architecture:
- TypeScript-based MCP server
- SQLite database for activity, lap, and health storage
- Direct integration with Garmin Connect API (via
garmin-connectlibrary) - Runs via
tsxfor development (no build step needed with Claude Desktop) - MCP SDK with stdio and Streamable HTTP transports
Database:
- Location:
data/garmin-data.db - Format: SQLite 3
- Updates: Incremental (only new activities downloaded)
- Contains: activity summaries, activity laps, and synced daily heart-rate summaries from Garmin Connect
- Node.js 20 or 22 (the versions exercised in CI)
- npm (comes with Node.js)
- Garmin Connect account with running/fitness activities
- An MCP-compatible client such as Claude Desktop, or an HTTP MCP client for self-hosting
- Garmin device that syncs to Garmin Connect (watch, bike computer, etc.)
Feel free to open issues or submit pull requests for:
- Bug fixes
- New data fields to track
- Additional MCP tools
- Documentation improvements
MIT
Note: This project is not affiliated with, endorsed by, or sponsored by Garmin Ltd. or Garmin International, Inc.