GARMIN Connect MCP

LLM-agnostic Garmin Connect MCP server for sleep and naps, recovery, Body Battery, training readiness, training load, running tolerance, VO2 Max, endurance and hill scores, race predictions, gear mileage, and running summaries.

Use the same local Garmin tools from any MCP client — Cursor, Continue, Codex CLI, Gemini CLI, Hermes, opencode — or from provider-backed demo agents using DeepSeek, OpenAI, OpenRouter, or Gemini.

Tribute & Thanks

GARMIN Connect MCP is built on kavakoss/mcp-garmin-connect. The original MCP server, Garmin client, caching layer, running tools, demo mode and test suite are the work of kavakoss. Huge thanks for building it and sharing it as open source under the MIT License. 🙏

This project extends that foundation with richer sleep data (naps, stages, overnight vitals), 20 new tools, and activity weather plus time in HR and power zones. kavakoss’s original commit history is preserved in this repository.

What’s New In This Version

• Richer get_sleep — naps (time, length, Garmin feedback), deep/light/REM/awake stages with ratings, bedtime and wake time, overnight HRV, SpO2, respiration, sleep stress, Body Battery recharge, skin temperature, and Garmin’s sleep need with recommended bedtime.
• 20 new tools — per-km splits from the raw recording, run/walk/stop segments, night-by-night sleep timeline, hour-by-hour day timeline, 90-day HRV trend, 52-week trends, daily activity, Body Battery, daily vitals, training readiness history, performance metrics, race prediction trend, running tolerance, saved workouts, yearly totals, badges, gear, devices, and body metrics (see MCP Tools).
• Richer existing tools — real HR and power zone ranges in get_zones, shoe usage in get_gear, watch alarms in get_devices.
• Richer get_activity_detail — weather during the activity (°C, km/h) and minutes in each heart-rate and power zone.
• Fix — get_recovery reports recovery time in hours (Garmin returns minutes).

See It In Action

Recorded with garmin-mcp serve --demo, which serves deterministic fictional sample data — no Garmin credentials and no real health data involved:

uv run garmin-mcp serve --demo

Works With Any MCP Harness

This server implements standard MCP over stdio and Streamable HTTP, so there is no harness-specific adapter to maintain. If your tool speaks MCP, it works:

Any client that supports MCP over stdio or Streamable HTTP can use every tool and prompt in this server with no extra code.

Why This Exists

Most fitness advice from LLMs is generic. This server lets an LLM inspect your real Garmin Connect data first, then answer with context from your activities, recovery markers, HRV, sleep, naps, Body Battery, stress, VO2 Max, and running history.

The core server is provider-neutral. DeepSeek, OpenAI, OpenRouter, and Gemini integrations are optional demo bridges for clients that do not speak MCP directly.

Features

• 36 Garmin tools and 4 prompt templates for training, recovery and sleep analysis.
• Demo mode — serve --demo runs on fictional sample data with no credentials, for try-outs and recordings.
• Local MCP server over stdio or Streamable HTTP (/mcp).
• Endpoint-level TTL cache with single-flight factories — concurrent tool calls share one Garmin request instead of stampeding the API.
• Global request concurrency cap so parallel snapshots stay within a safe request rate.
• Partial-error reporting — composite tools include an errors array when an endpoint fails, instead of silently returning nulls.
• Metric units — Garmin’s imperial weather values and gram-based weights are converted for you.
• Provider demo bridge for deepseek, openai, openrouter, and gemini.
• Explicit opt-in before Garmin health/activity data is sent to any external LLM provider.
• Mocked test suite plus optional live Garmin smoke tests.

How It Works

MCP (Model Context Protocol) is an open standard that lets AI assistants call local “tools”. This project is an MCP server: it runs on your own computer, logs in to Garmin Connect with your account, and exposes your data as tools your AI client can call when you ask a question.

Garmin watch ──sync──▶ Garmin Connect app ──▶ Garmin Connect (cloud)
                                                     │
                                         GARMIN Connect MCP (runs locally)
                                                     │  MCP tools
                                                     ▼
                                 Your AI client (Cursor, Codex CLI, Gemini CLI, opencode, …)

Your Garmin password and login tokens stay on your machine. The AI client only sees the tool results it asks for.

Getting Started (Step by Step)

1. Install the prerequisites

You do not need to install Python yourself; uv downloads a suitable version automatically.

2. Download the code

git clone https://github.com/sheloab/GARMIN-Connect-MCP
cd GARMIN-Connect-MCP
uv sync

3. Try it without a Garmin account (optional)

Demo mode serves fictional sample data, so you can check everything works before entering any credentials:

uv run garmin-mcp tools          # lists all tools and prompts
uv run garmin-mcp serve --demo   # starts the server on fictional data (Ctrl+C to stop)

To try demo mode inside your AI client, add "--demo" to the end of the args list in the client config shown in MCP Client Setup.

4. Add your Garmin credentials

Copy the example settings file and fill in your Garmin Connect email and password:

cp .env.example .env              # Windows PowerShell: Copy-Item .env.example .env

[email protected]
GARMIN_PASSWORD=your-password

.env is listed in .gitignore, so it is never committed. Leave the provider keys empty unless you plan to use the provider demo agents.

5. Log in once

uv run garmin-mcp login

Run this in a normal terminal: if your account uses two-factor authentication, Garmin sends a code and the command asks you to type it. The login tokens are saved in ~/.garminconnect (change with GARMIN_TOKEN_STORE), so later starts do not need to log in again.

6. Check the setup

uv run garmin-mcp doctor --live

Every required line should say OK. WARN lines for provider keys are fine if you do not use the provider demo agents.

7. Connect your AI client

Add the server to your client’s MCP configuration (see MCP Client Setup below), replacing /path/to/GARMIN-Connect-MCP with the folder you cloned. Restart the client, and the Garmin tools appear in its tool list.

You can also run the server yourself, over stdio or HTTP:

uv run garmin-mcp serve --transport stdio
uv run garmin-mcp serve --transport http --host 127.0.0.1 --port 8765
# HTTP clients connect to http://127.0.0.1:8765/mcp

8. Ask your first questions

Just ask in plain language; the AI picks the right tools. For example:

Keeping the data fresh

The server reads what is in Garmin Connect, not what is on your watch. Sync your watch in the Garmin Connect app before asking, otherwise you get the data from the last sync (get_devices shows the last upload time). Responses are cached for 2 minutes (GARMIN_CACHE_TTL_SECONDS), so after syncing, wait a moment before asking again.

MCP Client Setup

Generic stdio config for a local checkout:

{
  "mcpServers": {
    "garmin": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/GARMIN-Connect-MCP",
        "run",
        "garmin-mcp",
        "serve",
        "--transport",
        "stdio"
      ]
    }
  }
}

Credentials are read from the repo’s .env file, so no env block is needed.

opencode (opencode.json):

{
  "mcp": {
    "garmin": {
      "type": "local",
      "command": [
        "uv",
        "--directory",
        "/path/to/GARMIN-Connect-MCP",
        "run",
        "garmin-mcp",
        "serve",
        "--transport",
        "stdio"
      ],
      "enabled": true
    }
  }
}

Per-client examples (Cursor, Continue, Hermes, generic stdio/HTTP) live in docs/clients.

MCP Tools

Recovery and sleep

Activity and training

Performance

Gear, devices and body

MCP Prompts

• recovery_check
• weekly_training_review
• activity_analysis
• race_plan_context

Configuration

All settings come from environment variables (.env is loaded automatically).

Provider bridge variables (optional): DEEPSEEK_API_KEY, OPENAI_API_KEY, OPENROUTER_API_KEY, GEMINI_API_KEY, plus the matching *_BASE_URL and *_MODEL overrides.

How It Stays Fast And Quiet

• Endpoint-level cache. Sleeping, stress, RHR, and activity fetches are cached per day/limit and shared across every tool that needs them — a get_full_snapshot burst reuses the same responses instead of refetching.
• Single-flight. If two tool calls ask for the same missing key, only one hits Garmin; the other waits and reuses the result.
• Failures are never cached. A failed endpoint is retried on the next call.
• Concurrency cap. All outbound Garmin calls pass through one bounded semaphore (GARMIN_MAX_CONCURRENCY) so parallel sections cannot flood the API.
• Visible partial failures. When an optional endpoint fails, composite tools add an errors array (source + exception) to their output.

Provider Demo Agents

Provider API access may require a paid account or credits even if the provider’s web app has a free tier.

DEEPSEEK_API_KEY=sk-...
OPENAI_API_KEY=sk-...
OPENROUTER_API_KEY=sk-or-...
GEMINI_API_KEY=...

uv run garmin-mcp ask "How is my recovery today?" --provider deepseek --allow-external-health-data
uv run garmin-mcp ask "Summarize my running for 90 days." --provider openai --allow-external-health-data
uv run garmin-mcp ask "What should I focus on this week?" --provider gemini --allow-external-health-data

The --allow-external-health-data flag is required because Garmin tool results can be sent to the selected provider.

Privacy And Safety

• Your .env file and Garmin token cache stay local and are ignored by git.
• MCP clients that run locally can call Garmin tools without sending provider API keys to this project.
• Provider demo commands such as garmin-mcp ask can send Garmin health/activity results to the selected LLM provider, so the CLI requires --allow-external-health-data.
• get_zones returns only zone/threshold fields from the Garmin profile, not the raw profile (gender, birth date, weight, and similar PII are trimmed).
• Activity tools never return GPS coordinates, and workout/badge tools drop your Garmin display name and profile image, so routes and identity stay out of LLM conversations.

Testing

uv run garmin-mcp doctor
uv run garmin-mcp doctor --live
uv run garmin-mcp tools
uv run ruff check .
uv run pytest

CI runs on Linux (Python 3.11, 3.12, 3.13) and Windows (3.12). Optional live tests are gated and never run in CI:

$env:GARMIN_LIVE_TEST = "1"
uv run pytest tests/live

Troubleshooting

• Garmin rate-limited the request — lower GARMIN_MAX_CONCURRENCY (for example to 3) and retry after a few minutes.
• Garmin authentication failed — check GARMIN_EMAIL/GARMIN_PASSWORD, then run uv run garmin-mcp login from a real terminal (MFA prompts need interactive input).
• A tool returns errors — the array names the endpoint that failed; the rest of the payload is still valid partial data.
• A tool returns empty lists (for example get_body_metrics) — nothing was logged in Garmin Connect for that period.
• Today’s data looks old — sync the watch in the Garmin Connect app, wait a minute, and ask again; get_devices shows when the watch last uploaded.
• The AI client does not show the Garmin tools — check the --directory path in the client config points to your clone, then fully restart the client.

Garmin Disclaimer

This project uses the community garminconnect Python package and Garmin Connect endpoints. Garmin Connect is not a public API for individual open-source projects, and authentication or endpoint behavior may change without notice. This project is not affiliated with or endorsed by Garmin.

Some metrics depend on Garmin device capabilities. For example, a device may expose VO2 Max and Training Effect but not native Garmin Training Load or Training Status. Zone availability depends on what the Garmin user settings endpoint exposes for your account.

Contributing

Contributions are welcome. To add or change a tool:

1. Write the tool function in src/mcp_garmin_connect/tools.py. Fetch Garmin data through _endpoint(...) so it is cached and failures are reported in an errors array instead of crashing.
2. Register it in TOOL_SPECS at the bottom of the same file, with a clear docstring (it becomes the tool description the AI reads) and a JSON schema for its parameters.
3. Add a test in tests/test_tools.py using the FakeGarmin client with fictional values, and update the tool count in tests/test_server.py.
4. Run uv run ruff check . and uv run pytest before opening a pull request.

Please keep personal data out of contributions: no real Garmin IDs, GPS coordinates, health values or credentials in code, tests or issues.

License

This project is licensed under the MIT License.

Original work © 2026 kavakoss — kavakoss/mcp-garmin-connect. Extensions © 2026 sheloab.