garmin-mcp
Enables users to analyze their own Garmin Connect data—activities, sleep, HRV, Body Battery, training readiness—directly inside Claude Desktop.
README
Garmin → Claude MCP server
Analyze your own Garmin Connect data — activities, sleep, HRV, Body Battery, training readiness — right inside Claude Desktop.
A fully local, read-only MCP server that lets you analyze your own Garmin Connect data in Claude Desktop. The server runs as a local child process of Claude Desktop and talks to it over stdio.
Guiding principles
- 100% local — nothing is hosted or exposed to the network (stdio).
- Read-only — there is no write endpoint to Garmin.
- No third party — the only network contact is Garmin. No telemetry, no external logging.
- Passwordless at runtime — the password is never stored; after the one-time login only local tokens are used.
Requirements
- Node.js ≥ 20 (tested with Node 22 LTS), npm.
- A Garmin Connect account.
Setup (4 steps)
# 1) Install dependencies
npm install
# 2) Milestone test: does the client get through Garmin's Cloudflare protection?
npm run smoke
# Expectation: "Status 200" and "Challenge: no".
# 3) One-time interactive login (prompts for email, hidden password, MFA if enabled)
npm run login
# Creates ~/.garmin-mcp/tokens.json (mode 0600). The password is NOT stored.
# 4) Compile TypeScript to build/
npm run build
Connecting Claude Desktop
Claude Desktop → Settings → Developer → Edit Config. Add the server with its
absolute path (claude_desktop_config.json):
{
"mcpServers": {
"garmin": {
"command": "node",
"args": ["/absolute/path/to/garmin-mcp/build/mcp/server.js"]
}
}
}
Then restart Claude Desktop. Try it in chat, e.g.:
"Am I logged in to Garmin? Use
whoami." "Show me my daily summary and my recent activities."
Available tools
13 read-only tools. Date parameters are YYYY-MM-DD (default = today); ranges
default to the last ~4 weeks. Several tools take a metrics[] / include[] selector,
so one tool covers many data types (this keeps the tool list small for good tool
selection). Long time series in responses are truncated to stay compact.
| Tool | Description |
|---|---|
whoami |
Connection check + account profile |
get_daily_health |
Daily wellness for a date — metrics[]: summary, sleep, stress, heart_rate, hrv, spo2, respiration, hydration, steps, floors, intensity_minutes, body_battery, body_battery_events, stats_and_body |
get_training |
Training for a date — metrics[]: readiness, morning_readiness, status, vo2max, fitness_age |
get_fitness |
Fitness/performance — metrics[]: race_predictions, cycling_ftp, lactate_threshold, personal_records, endurance_score, hill_score, resting_heart_rate, weekly_intensity_minutes (startDate/endDate) |
get_weight |
Body weight & composition over a range (include_raw also returns weigh-ins) |
get_steps_history |
Step totals over a range — granularity: daily or weekly |
get_activity |
One activity by activityId — include[]: summary, splits, weather, details, hr_zones, exercise_sets |
list_activities |
Recent activities, or by date range/type (limit/startDate/endDate/type) |
get_devices |
Paired Garmin devices |
get_user_profile |
User profile & settings (units, preferences) |
get_goals |
Goals (status: active/future/past) |
get_workouts |
Saved workouts (limit) |
get_scheduled_workouts |
Scheduled workouts / calendar (year/month) |
How it works
The server registers each Garmin data method as an MCP tool and talks to Claude
Desktop over stdio. Every call goes through a generic connectapi() request to
connectapi.garmin.com with a bearer token, sent via
cycletls so the TLS fingerprint looks
like a real browser (Garmin sits behind Cloudflare).
Authentication happens once (npm run login): a login cascade — mobile iOS JSON
login first, the classic widget/CSRF flow as fallback — yields a CAS service ticket,
which is exchanged in Garmin's DI-OAuth2 flow for an access + refresh token. The
tokens are cached locally and the access token is refreshed automatically on expiry;
your password is never stored.
Security model
- The password is never stored — it is only used for the one-time login.
- Tokens live at
~/.garmin-mcp/tokens.jsonwith file mode0600(only you can read them). - The access token is renewed automatically via the refresh token when it expires.
- Read-only: there is no code path that changes anything on Garmin.
- No telemetry, no external logging. Diagnostics go to
stderronly (never over the stdio MCP channel).
Architecture (layers)
src/http/impersonate.ts— TLS impersonation viacycletls(JA3 fingerprint) + cookie jar, so Garmin's bot protection lets us through.src/http/smoke.ts— milestone test againstsso/embed.src/garmin/auth.ts— login cascade (mobile iOS JSON login → widget/CSRF fallback) + MFA + DI-OAuth2 ticket exchange + refresh.src/garmin/tokens.ts— local token cache (0600).src/garmin/client.ts— genericconnectapi()+ typed, read-only data methods.src/mcp/server.ts— MCP server, tool registration, stdio transport.src/cli/login.ts— one-time interactive login.
Maintenance & risks (named honestly)
This uses Garmin's unofficial internal API, ported from the open-source Python
package python-garminconnect.
Expected breakage points and how to fix them:
- Unofficial API — endpoint paths can change. On errors, compare the paths in
src/garmin/client.tsagainst the currentpython-garminconnectsource. - JA3 / User-Agent — on a
403or a "Just a moment …" Cloudflare page, update the JA3/User-Agent pair insrc/http/impersonate.tsto a current Chrome (both together). - Client ids — the DI-OAuth2 client ids (
GARMIN_CONNECT_MOBILE_ANDROID_DI_*) rotate; update the list insrc/garmin/auth.tson auth errors. garthis discontinued — the previously common auth librarygarthis no longer maintained; the authoritative reference ispython-garminconnect.- Keep the token folder private —
~/.garmin-mcp/holds valid session tokens; don't share it, don't commit it. - Terms of service — access via the unofficial API may be in tension with Garmin's terms of service. Intended for private personal use with your own data.
Troubleshooting
403or a "Just a moment …" page — Garmin's Cloudflare is blocking the TLS fingerprint. Update the JA3/User-Agent pair insrc/http/impersonate.tsto a current Chrome (both together), then re-runnpm run smoke.🔒 Not logged in/ session expired — runnpm run loginagain.⏳rate limited (HTTP 429) — Garmin is throttling; wait a bit and retry.- First run is slow / cycletls errors —
cycletlsdownloads a small Go helper binary on first use; make sure it can execute and isn't blocked by the OS.
Re-login
If tools report 🔒 Not logged in (token expired/invalid):
npm run login
License
MIT © Marcel Fortmann
Disclaimer
This is an unofficial, community project. It is not affiliated with, endorsed by, or sponsored by Garmin. Garmin® and Garmin Connect™ are trademarks of Garmin Ltd. or its subsidiaries.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.