Garmin-TN-MCP (garmin-raw)

English | Русский

Minimal raw access to Garmin Connect for training analysis. One backend (garminconnect 0.3.x) powers two frontends:

• MCP server (garmin-raw-mcp) — live data access inside Claude Desktop chat.
• One-shot export (garmin-raw-export) — dumps a date range to JSON for

reuse with other athletes (the file is uploaded into a chat).

Principle: raw data only (per-lap HR / cadence / power / stride / elevation, per-second streams, the lactate comment). No VO2max / training-effect / device threshold estimates — those are deliberately excluded.

Install

Requires Python 3.10+ and uv.

git clone https://github.com/asilenin/Garmin-TN-MCP.git
cd Garmin-TN-MCP
uv sync

1. Authenticate (once)

uv run garmin-raw-auth

Enter email, password and the MFA code. Tokens are saved to ~/.garminconnect (0.3.x format). After that no login is needed — the server and the export run on tokens (and avoid 429 rate-limiting from repeated logins).

2. Connect the MCP to Claude Desktop — one command

uv run garmin-raw-install

It auto-resolves uv and this folder's path and safely merges the garmin-raw server into claude_desktop_config.json without overwriting the rest (your preferences, etc.), with a backup. Cross-platform (macOS/Windows/Linux). You can pass the path explicitly:

uv run garmin-raw-install /full/path/to/Garmin-TN-MCP

Then fully restart Claude Desktop (Cmd+Q on macOS) — the 6 tools appear.

<details> <summary>Manual alternative (if you prefer not to use the script)</summary>

Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "garmin-raw": {
      "command": "/Users/<you>/.local/bin/uv",
      "args": ["--directory", "/full/path/to/Garmin-TN-MCP", "run", "garmin-raw-mcp"]
    }
  }
}

</details>

Uninstall

uv run garmin-raw-uninstall   # removes garmin-raw from the config (with backup), leaves the rest

Restart Claude Desktop. Tokens are not removed — delete them manually if you want:

rm -rf ~/.garminconnect        # saved Garmin tokens

Remove the whole thing:

cd .. && rm -rf Garmin-TN-MCP  # the repository itself

Export (for reuse with other athletes)

# whole period
uv run garmin-raw-export --start 2026-06-01 --end 2026-06-21

# single activity + per-second streams
uv run garmin-raw-export --start 2026-06-20 --end 2026-06-21 \
    --activity 23321211303 --streams

Produces garmin_export.json — upload it into a chat.

Tools (identical in the MCP and the export)

Lactate

Write it into the activity comment in Garmin Connect as LA:6.1 (context is fine: LA:6.6 @rep12). get_activity_comment reads the description field and parses every value into lactate_mmol. The comment is fetched lazily — only for activities actually under analysis — to avoid doubling the request count.

Notes

• Garmin PRs are auto-detected fastest splits, not certified race times; they

can be faster than official results. Use certified times as form markers and Garmin PRs only as a hint.

• Wellness/PR methods are resolved by trying candidate names: if your

garminconnect version renames one, the tool returns _error instead of crashing the whole response.

• PII (owner name/ID) is stripped from outputs, case-insensitively — hygiene

for shared exports.

• If the MCP silently stops responding, it's almost always stale tokens: re-run

garmin-raw-auth. The one-shot export is the robust fallback.

Garmin disclaimer

This project accesses Garmin Connect through an unofficial method (the community python-garminconnect library, which logs in with your own credentials). It is not affiliated with, endorsed by, or supported by Garmin. Your use may be subject to Garmin's Terms of Service; you use it at your own risk. No warranty is provided (see LICENSE).

Authorship & AI generation

This project was designed and written by Claude (Anthropic's AI assistant) during an extended pair-programming session, under the direction, review and testing of Anton Silenin. The methodology, architecture, debugging and final verification against real Garmin data were done collaboratively in conversation: the human author initiated the work, made the design decisions, validated every step on live data, and is the copyright holder.

AI-generated output carries no separate human authorship under copyright law, so it is released under the human author's name (MIT, see LICENSE). This note is here for transparency, not as a license requirement. As with any AI-assisted code, review it before relying on it.