Frequently asked questions

A hosted Garmin Connect MCP server: you connect your Garmin once, paste one URL into your AI assistant, and it can read your health and training data and, on higher plans, write workouts to your watch.

ChatGPT (paid plans, developer mode), Claude (all plans including free), Claude Code, Cursor, Windsurf, Zed, Cline, Perplexity Pro, Mistral Le Chat and any client that supports remote MCP servers over Streamable HTTP. Guides: ChatGPT, Claude, developer clients.

Yes. Once the connector is added to your ChatGPT or Claude account it works in their mobile apps too. Nothing runs on your phone.

Any device that syncs to Garmin Connect: Forerunner, Fenix, Epix, Venu, Vivoactive, Instinct, Enduro, Edge, Index scales, and so on. We read Garmin Connect, not the watch, so if it's in the app it's available.

No. Garmin Community MCP is independent and not affiliated with, endorsed by, or sponsored by Garmin Ltd., OpenAI or Anthropic.

garminmcp.com is a read-only connector with about ten tools; Garmin Community MCP has 50+ data sources and writes workouts to your watch. They are unrelated services. See the comparison.

Sleep, HRV, stress, Body Battery, resting and all-day heart rate, respiration, SpO2, steps, floors, intensity minutes, hydration, menstrual cycle, weigh-ins, body composition, blood pressure, training readiness, training status, VO2 max, race predictions, endurance and hill scores, lactate threshold, personal records, badges, goals, activities with splits, HR zones, time series and weather, strength sets, workouts, the training calendar, training plans, nutrition logs, gear and devices. Anything else is reachable with the raw garmin_api_get tool.

As far as Garmin Connect holds: years. Daily endpoints take a date; activity lists are paginated. "Analyse my last 12 months" works, it just takes more tool calls.

It's as fresh as your last sync to Garmin Connect. Sync your watch and the AI sees it within seconds.

It can. daily_overview trims long time series to keep answers quick; get_garmin_data returns the full arrays when the model asks for them.

No. Data passes through our server to your AI client and is not written to disk. We cache a short-lived Garmin access token and keep counts of tool calls for rate limiting. See privacy.

On Athlete and Coach, yes: create, schedule and update workouts; log weight, hydration and blood pressure; rename or retype activities; log manual activities; link gear. Permanent deletes are off by default and must be enabled in the dashboard. Read has no write tools.

Every write tool is annotated as non-read-only, and ChatGPT and Claude both ask before running such tools. You can additionally turn individual tools off in your client, or switch writes off entirely in our dashboard.

The AI creates a structured workout in your Garmin Connect library and, if you asked, schedules it on the calendar. Your watch picks it up at the next sync, like any workout you build in the Garmin app.

Running, cycling, swimming, strength, cardio, HIIT, yoga, pilates, mobility, multisport and other. Steps by time, distance, lap button, calories or reps; targets by pace, heart-rate zone or bpm, power zone or watts, cadence or speed. Strength steps can name Garmin exercises.

No. You sign in through our page; the password goes to Garmin's sign-in service and is not written anywhere. We store an encrypted long-lived token, which you can revoke from the dashboard or by changing your Garmin password.

Garmin's official Health API is only available to approved businesses, and new applications have been paused since spring 2026. We use the same interface the Garmin Connect mobile app uses, with the authentication flow pioneered by the open-source garth and garmin-connect libraries. It exposes everything the app can show and can create workouts, which the official API cannot.

We make requests at the pace of a human conversation using the app's own interface. Community libraries have done this for years without reports of account action for personal use. We can't speak for Garmin, so our terms disclose the risk, there's no long-term lock-in, and annual plans have a 14-day refund.

We monitor error rates and usually patch within hours. Incidents are posted on the status page. The raw API tool keeps working for anything not yet re-catalogued.

Yes: "Disconnect Garmin" in the dashboard deletes the token immediately. Changing your Garmin password also invalidates it. Rotating your connector token cuts off any AI client that had the old one.

48 hours from the moment you connect your Garmin account, with full Athlete features. A card is required to start, but nothing is charged if you cancel before the trial ends, and we email you 8 hours before. If you connected but barely used it, we extend the trial automatically.

It keeps trials to one per person and means your connector keeps working on day three without a gap. You are not charged unless the trial ends with the subscription still active.

Prices are listed in USD. Stripe shows and charges your local currency at checkout. Garmin Community MCP is an Australian company; Australian customers pay GST on top.

Setup is five minutes and the data arrives nightly, so two days shows you the real thing. Longer trials mostly produce forgotten signups. We're testing this; if a week converts more happy users we'll change it.

From the dashboard, in one click, via the Stripe billing portal. Your plan stays active to the end of the paid period. You can also delete your account, which removes your Garmin token and all data within 24 hours.

Annual plans: 14 days, no questions. Monthly: cancel any time, no partial refunds.

Claude: no, the free plan allows one custom connector. ChatGPT: yes, custom connectors need developer mode on a paid plan.

Your Garmin token expired (they last about a year) or you changed your Garmin password. Reconnect from the dashboard.

Check the connector is enabled for the current chat, and tell the model to use the Garmin tools. In ChatGPT, confirm developer mode is still on.

Set your time zone in the dashboard.

Read is 300/day, Athlete 1,000/day. Deep multi-month analyses can use a lot of calls; ask the model to use daily_overview and summaries, or upgrade.