Garmin MCP

A remote MCP server that exposes your Garmin Connect data — activities, sleep, heart rate, training metrics, workouts, nutrition — to AI clients like Claude and ChatGPT. Runs on the Cloudflare Workers free tier. A TypeScript port of taxuspt/garmin_mcp.

View on GitHub Open interactive diagram ↗ Connect a client
143 tools + 5 resources OAuth 2.1 TypeScript Free tier

What it is

One remote endpoint your AI client talks to; it relays to Garmin on your behalf.

Garmin offers no public API for personal data, so this server speaks Garmin Connect's internal API — the same one the mobile app uses — through a hardened, single-purpose Cloudflare Worker. You authenticate once in the browser; the server exchanges your long-lived Garmin token for short-lived bearers on each call and returns clean JSON to the model.

How it works

The one-time authorize flow and the per-request tool call, end to end.

Sequence diagram: one-time authorize flow and per-request tool call between MCP client, Worker, KV, GarminMCP, and Garmin Connect Open interactive diagram ↗

The interactive version has light/dark, pan/zoom, search, and guided views.

Tools

143 tools across nine modules, plus 5 workout-template resources. Full parity with the Python original, adapted for a serverless host. 98 tools declare typed output schemas and return structured content.

30
Health & wellness
Sleep (per-night + bulk range), HR, stress, SpO2, body battery, steps, respiration — full and lightweight summaries.
24
Activity management
List, detail, splits, weather, gear; rename, retype, RPE, manual entries, file upload, HR backfill, delete.
19
Workouts & builders
CRUD, scheduling, GraphQL calendar, plus run/walk/strength builders.
18
Challenges · devices · gear
Badges, records, challenges, device settings/alarms, gear stats.
15
Training & performance
Training status/load, HRV, VO2max, endurance & hill score, trends.
14
Nutrition
Food logs, meals, custom foods, macro goals, quick-add logging.
14
Weight · body · courses
Weigh-ins, body composition (FIT upload), blood pressure, hydration, women's health, GPX courses.
5
Profile & passthrough
Profile/settings plus garmin_get, a raw GET escape hatch.
4
FIT analysis
Power-duration curve, normalized power, W/kg from downloaded FIT files.

Deploy your own

Cloudflare Workers free tier. Roughly ten minutes.

git clone https://github.com/rohankmr414/garmin-mcp
cd garmin-mcp
npm install

# 1. create the KV namespace, paste the id into wrangler.jsonc
npx wrangler kv namespace create OAUTH_KV

# 2. deploy
npx wrangler deploy

# 3. lock it to your Garmin account (displayName from get_user_profile)
npx wrangler secret put OWNER_GARMIN_IDS

Set OWNER_GARMIN_IDS to a comma-separated allowlist of Garmin displayNames. Leave it empty to allow any Garmin account.

Connect a client

Add the server URL; the browser walks you through a one-time Garmin login.

Claude Code

claude mcp add garmin --transport http \
  https://garmin-mcp.<your-subdomain>.workers.dev/mcp

ChatGPT (developer mode)

Settings → Connectors → Create. Enter the same /mcp URL, choose OAuth authentication, and leave scopes empty. ChatGPT registers itself and opens the authorize page.

claude.ai web

Settings → Connectors → Add custom connector, with the same URL.

On the authorize page you enter your Garmin email, password, and MFA code (or paste a token). Credentials are used once to log in, then discarded — only the resulting token is stored, encrypted.

Security model