Garmin Workouts MCP

Garmin Workouts MCP

garmin-workouts-mcp is a standalone MCP server for Garmin Connect workouts.

It is intended as a focused extension for workflows that need a bit more structure around Garmin workout payloads, especially strength training.

This project is packaged as a stdio MCP server and can be published as an OCI image for MCP registries and Glama deployment. It is not a standalone public HTTP MCP endpoint.

Additions

• Supports Garmin strength workout steps with reps end conditions.
• Supports exercise metadata via explicit Garmin enums or friendly aliases.
• Adds preview_workout_payload so payloads can be inspected before upload.
• Adds validate_workout for early schema and mapping errors.
• Adds resolve_supported_strength_exercise for quick mapping checks.
• Adds get_workout_input_schema for machine-readable client integration.
• Includes walking as a supported sport type, which is also reflected in the prompt/schema.
• Keeps the familiar list/get/delete/schedule/calendar/activity tools.

Environment

Garmin-backed tools authenticate lazily when they are called:

• Authentication path: GARMIN_EMAIL and GARMIN_PASSWORD

The server can start without credentials. Tools that do not talk to Garmin, such as payload preview and schema inspection, still work without secrets.

Workout Input

The upload and preview tools accept a JSON object shaped like this:

{  "name": "Upper Day",  "type": "strength",  "steps": [    {      "stepType": "warmup",      "endConditionType": "lap.button",      "stepDescription": "General warm-up"    },    {      "stepType": "interval",      "exercise": "incline db press",      "endConditionType": "reps",      "stepReps": 8,      "stepDescription": "8-10 reps"    },    {      "stepType": "rest",      "endConditionType": "time",      "stepDuration": 120    }  ]}

For strength exercises, either pass a friendly alias:

{ "exercise": "t bar row" }

or explicit Garmin enums:

{  "exercise": {    "category": "ROW",    "exerciseName": "T_BAR_ROW"  }}

You can also inspect the accepted structure programmatically through get_workout_input_schema, or resolve likely Garmin strength mappings with resolve_supported_strength_exercise.

Development

Run tests in Docker Compose:

docker compose run --rm tests

Build the runtime image:

docker build -t garmin-workouts-mcp:local .

Smoke test the stdio server startup without Garmin credentials:

python - <<'PY'import subprocess
proc = subprocess.Popen(    ["bash", "-lc", "tail -f /dev/null | docker run --rm -i garmin-workouts-mcp:local"])try:    proc.wait(timeout=5)    print(f"container exited early with code {proc.returncode}")finally:    if proc.poll() is None:        proc.terminate()        proc.wait()        print("container stayed up for 5 seconds")PY

Publishing

The intended OCI image location is:

ghcr.io/pranciskus/garmin-workouts-mcp

Registry metadata lives in server.json. The OCI image carries the required label:

io.modelcontextprotocol.server.name=io.github.pranciskus/garmin-workouts-mcp

Glama ownership metadata lives in glama.json. It declares the GitHub maintainer account that can claim and manage the Glama listing.