Garmin Personal Coach

Languages: English | 한국어 | Español | 日本語

Garmin Personal Coach is a Garmin-first AI coaching engine for endurance athletes.

It works today through:

• CLI
• Telegram
• MCP / OpenClaw

It is not yet a dashboard, mobile app, or food-photo calorie app.

• Connects to Garmin Connect as the primary source of truth
• Supports Strava as a supplemental sync source
• Calculates training load (CTL / ATL / TSB)
• Gives coaching guidance through CLI, Telegram, and MCP/OpenClaw
• Supports optional AI providers: OpenAI, Anthropic, Gemini
• Supports lightweight personalized nutrition coaching

Recommended:

pip install garmin-personal-coach[all]

From source:

git clone https://github.com/HOYALIM/garmin-personal-coach.git
cd garmin-personal-coach
pip install -e .[all]

garth login your@email.com

garmin-coach setup

Create a Strava API app, then run:

garmin-coach connect-strava

Strava is supplemental only in this release. Garmin remains the primary source of truth.

To inspect sync state:

garmin-coach oauth-status
garmin-coach strava-sync --dry-run

garmin-coach status
garmin-coach log
garmin-coach oauth-status
garmin-coach garmin-sync --dry-run
garmin-coach strava-sync --dry-run
garmin-coach --version
garmin-coach --check-updates

export TELEGRAM_BOT_TOKEN="your_bot_token"
garmin-coach-telegram

For a bot token:

1. Open Telegram
2. Talk to @BotFather
3. Create a new bot
4. Copy the token into TELEGRAM_BOT_TOKEN

Common commands:

• /start
• /status
• /plan
• /help

Natural-language examples:

• How is my condition today?
• I finished my workout
• What should I do tomorrow?

Recommended MCP config:

{
  "mcpServers": {
    "garmin-coach": {
      "command": "garmin-coach-mcp",
      "args": []
    }
  }
}

Fallback:

{
  "mcpServers": {
    "garmin-coach": {
      "command": "python",
      "args": ["-m", "mcp_server"]
    }
  }
}

Available MCP tools:

• get_training_status
• get_user_profile
• get_recent_activities
• handle_natural_language
• health
• get_training_plan

Optional AI enhancement:

export OPENAI_API_KEY="sk-..."
# or
export ANTHROPIC_API_KEY="sk-ant-..."
# or
export GEMINI_API_KEY="..."

Without API keys, the product still works with rule-based coaching.

The current release supports lightweight personalized nutrition coaching based on:

• training load / fatigue context
• weight goal (maintain, lose, gain)
• dietary style (omnivore, vegetarian, vegan, other)
• food restrictions / avoidances
• preferred coaching style (brief, detailed, macros)

This is guidance only in the current release.

Not included yet:

• meal logging
• barcode scanning
• image upload
• calorie estimation from photos

In a future update, the nutrition layer may expand to image-based meal analysis where a user uploads a meal photo and gets calorie estimation plus meal recommendations.

That is not part of the current release.

• Garmin-first coaching engine
• Strava supplemental sync
• CLI / Telegram / MCP usage
• lightweight nutrition coaching

• Web dashboard
• iMessage integration
• Nike Run Club integration
• Apple HealthKit / Apple Watch integration
• full meal tracking or photo-calorie workflows

garmin_coach/
├── adapters/          # Garmin/Strava data access, Nike scaffold only
├── handler/           # Natural language coaching core
├── integrations/      # Garmin / Strava sync-to-load flows
├── wizard/            # Interactive setup
├── nutrition/         # Nutrition guidance logic
├── telegram_bot.py    # Telegram runtime
└── cli.py             # CLI entrypoint

mcp_server/
├── server.py          # MCP handlers
└── entrypoint.py      # MCP stdio entrypoint

• Main profile/config: ~/.config/garmin_coach/config.yaml
• Strava token: ~/.config/garmin_coach/strava_token.json

This release should be understood as an honest beta for local/power users:

• Garmin-first
• usable now through CLI, Telegram, and MCP/OpenClaw
• still early for broader consumer UX

MIT