FAQ

Is my data sent to an LLM provider?

The server sends nothing anywhere on its own: it talks to Garmin Connect to fill the mirror, and answers MCP clients that connect to it. What those clients do with the answers is up to the client. When a client such as Claude Code, ChatGPT or Cursor calls a tool, the result goes into the conversation and therefore to whatever model that client uses, hosted or local. Narrow what a client can see with the tool profiles and GARMIN_DISABLED_TOOLS, and treat the mirror, the snapshot and the FIT files as the health data they are: never attach them to an issue.

Can this get my Garmin account locked or rate-limited?

The design is built around not doing that. Requests go out one at a time, about 1.5 s apart, a 429 pauses all background work for 30 minutes, and the server never logs in on its own: you log in once with garmin-mcp-auth (with the MFA code), and python-garminconnect refreshes the tokens in place for months. Repeated logins are what gets an address rate-limited or an account locked, which is why GARMIN_EMAIL/GARMIN_PASSWORD are a one-shot fallback and never retried. If another client also polls your account (a second instance, a third-party app), the budget is shared: see Troubleshooting.

Can I mirror several accounts?

One account per server: one mirror, one token store, one process. Run a second instance (another compose project with its own .env, volume and port, or a second GARMIN_DB_PATH and GARMINTOKENS outside Docker) and give each client the server it should talk to.

Garmin Connect China?

Set GARMIN_IS_CN=true before running garmin-mcp-auth and keep it set for the server; tokens from garmin.com and garmin.cn are not interchangeable, and a mismatch is reported as region_mismatch.

How long does the first sync take?

The profile, devices and the activity list come first, then the per-range daily summaries, so the recent months are there within hours. Each day's datasets and each activity's detail and FIT file follow, newest first, at about one request every 1.5 s: a few years of history take about a week. Tools work from the first minute and fetch on demand what the mirror does not have yet. Garmin's account export supplies every FIT file at once and shortens a long backfill to hours; GARMIN_HISTORY_START limits how far back the mirror goes.

Does it write to Garmin?

Only when you ask it to. The sync only reads. Writes happen through the write tools you call (renaming or deleting an activity, uploading a workout, logging weight or hydration, and so on; the tools reference marks each as read, write or destructive) and, if you enable it, the optional Strava sync, which renames activities and deletes orphaned Strava-made courses within documented caps and a dry-run default.

Where is my data, and how do I delete it?

In Docker, everything is in the garmin-data volume (/data): the mirror, the snapshot, the FIT files and the tokens. docker compose down -v removes the containers and the volume; docker image rm ghcr.io/justmytwospence/garmin-mcp the image. Outside Docker, delete the data directory (~/.local/share/garmin-mcp, ~/Library/Application Support/garmin-mcp, or wherever GARMIN_DB_PATH pointed) and ~/.garminconnect (the tokens), then uv tool uninstall garmin-mirror-mcp. Nothing is stored anywhere else. Your data on Garmin Connect is untouched either way.

Does it run on arm64, a Raspberry Pi, a NAS?

The image is built for amd64 and arm64, so a Raspberry Pi 4 or 5 (64-bit OS), an Apple Silicon Mac and most ARM servers run it. The compose file caps the container at 1.5 GB and DuckDB at 512 MB; a machine with 2 GB of RAM is enough for the server alone, the Grafana and notebook profiles want more. See Memory.

Does the mirror replace a backup of my Garmin data?

It is a faithful local copy of what Garmin returned (every response verbatim, plus the original FIT files), which is more than Garmin's own export contains for most datasets, and the backup profile archives it nightly. It is still a copy of a live account: the snapshot is the file to keep.

Which tools does it have, and are they the same as upstream?

160 tools: the 157 of Taxuspt/garmin_mcp, with the same names, parameters and descriptions, answered from the mirror, plus garmin_query, garmin_schema and garmin_sync_status. The Strava sync adds 3 more when it is enabled. The tools reference lists them all.