Garmin data for LLMs, minus the API archaeology

What it does

This Model Context Protocol server is a thick adapter between Garmin Connect and AI assistants like Claude. It wraps the python-garminconnect library into more than 110 callable tools, letting an LLM inspect your runs, check sleep and stress, edit activity details, download raw FIT files, and schedule workouts on your watch.

The interesting bit

The real value is in the high-level workout builders. Instead of hand-crafting Garmin’s finicky JSON—with its numeric conditionTypeId mappings and pace values in meters per second—you can ask the AI to generate an interval run or strength session and the server produces a valid payload. It also surfaces deep cycling analytics like power duration curves and DI2 shift intelligence, provided you own the hardware.

Key highlights

• 110+ tools covering activities, health metrics, body composition, gear, nutrition, training load, and women’s health
• Read-write access: edit activity names, types, descriptions, RPE, and perceived feel
• Workout builders for walk/run intervals, Z2 walks, and strength sessions that abstract Garmin’s raw JSON schema
• Download activity files in FIT, GPX, TCX, or CSV formats
• Tool filtering via allowlist/denylist environment variables to keep LLM context windows manageable

Caveats

• Advanced activity analysis (power zones, FIT parsing, DI2 data) requires compatible power meters or electronic shifting hardware
• Several endpoints are deliberately omitted, including large GPS track downloads, destructive deletes, and sport-specific workout uploads
• With 110+ tools registered by default, the server can bloat an LLM’s system prompt unless you filter them

Verdict

Useful if you want to query, edit, or build workouts from your Garmin data through Claude or another MCP client. Not for you if you don’t use Garmin Connect or prefer to keep your training logs out of LLM context windows.