fitbit-mcp

MCP server for the Fitbit Web API with OAuth PKCE, local SQLite cache, and trend analysis.

Designed for Claude Code and other MCP clients. Syncs your Fitbit data to a local database for fast, offline queries - no API calls needed after the initial sync.

Features

• OAuth 2.0 PKCE - Secure auth flow, no client secret needed
• Local SQLite cache - Sync once, query instantly
• Incremental sync - Only fetches new data since last sync
• MCP tools - Sync, query (cached data types plus live-only devices/lifetime/goals), and trend analysis
• Live mode - Bypass cache and query the API directly
• CLI - Auth setup, sync, and JSON import from the command line
• Rate limit handling - Activity and food log syncs sleep and retry on 429; other data types report rate_limited and resume on the next sync

Data types

Requirements

• Python 3.13+ (tested on 3.13 and 3.14 in CI)
• A Fitbit developer account with a registered personal app

Setup

1. Install

pip install fitbit-mcp

Or run it without installing:

uvx fitbit-mcp

For development from a clone:

pip install -e ".[dev]"

2. Register a Fitbit app

1. Go to dev.fitbit.com/apps and create a new app
2. Set OAuth 2.0 Application Type to Personal
3. Set Redirect URL to http://localhost:8080
4. Note your Client ID (you won't need the client secret - PKCE doesn't use one)

3. Authenticate

fitbit-mcp auth

This opens your browser for Fitbit login, exchanges the auth code via PKCE, and saves tokens locally.

Tokens are stored in ~/.config/fitbit-mcp/fitbit_tokens.json with 0600 permissions. Access tokens expire in 8 hours and are refreshed automatically. Refresh tokens expire after 90 days of inactivity.

4. Register with Claude Code

claude mcp add -s user fitbit -- fitbit-mcp

5. First sync (optional)

Query tools auto-sync on first use, so you can skip this step. To pre-populate the cache or sync a longer history, run:

fitbit-mcp sync --days 30

CLI usage

fitbit-mcp              Start the MCP server (stdio transport)
fitbit-mcp -V, --version  Print the installed package version
fitbit-mcp auth         Interactive OAuth setup
fitbit-mcp sync         Sync data to local cache
  --days N              Days of history for first sync (default: 30)
  --types TYPE,...      Data types to sync (default: all)
  --since YYYY-MM-DD    Backfill from this date, overriding the incremental
                        resume-from-last-sync cursor and --days
  --until YYYY-MM-DD    Inclusive end date for a --since backfill; together
                        they re-fetch exactly that window (e.g. to repair a
                        gap in the middle of the cache)
fitbit-mcp import       Import existing JSON data files
  --data-dir PATH       Directory containing JSON files

MCP tool reference

Query tools auto-sync on the first query of each day per data type. Use live=True to bypass the cache entirely and fetch directly from the API.

All query tools accept these common parameters:

• start_date - Start date as YYYY-MM-DD, YYYY-MM, or 30d (relative). Default: last 30 days.
• end_date - End date as YYYY-MM-DD. Default: today.
• live - If true, fetch from Fitbit API instead of cache (bypasses auto-sync).

fitbit_get_exercises also accepts:

• exercise_type - Filter by activity name (case-insensitive substring match), e.g. "cycling", "walk", "run". Default: all types.

fitbit_sync

Syncs data from the Fitbit API to the local SQLite cache. Query tools call this automatically on first use of the day, so explicit calls are only needed for longer history or forced refresh.

• data_types - What to sync: all, heart_rate, activity, exercises, sleep, weight, spo2, hrv, azm, breathing_rate, skin_temperature, core_temperature, cardio_fitness, food_log. Comma-separated. Default: all.
• days - Days of history for first sync (default: 30). Subsequent syncs are incremental.
• since - Optional YYYY-MM-DD. Backfill from this date regardless of what is already cached, overriding incremental resume and days.
• until - Optional YYYY-MM-DD inclusive end date; requires since. Together they re-fetch and upsert exactly the since..until window - use to repair a gap in the middle of the cache without re-pulling everything up to today.

fitbit_trends

Aggregated trend analysis from cached data.

• data_type - What to analyse: heart_rate, activity, exercises, sleep, weight, spo2, hrv, azm, breathing_rate, skin_temperature, core_temperature, cardio_fitness, food_log. Default: activity.
• period - Aggregation: weekly, monthly, quarterly. Default: monthly.
• start_date - Start date. Default: last 12 months (365 days).
• end_date - End date. Default: today.
• compare - Compare two periods: last_30d vs previous_30d, 2026-03 vs 2026-02, 2026-Q1 vs 2025-Q4. When set, period/start_date/end_date are ignored.

OAuth scopes

The following Fitbit API scopes are requested during setup:

These are the scopes needed for all tools. If you only need a subset, edit FITBIT_SCOPES in config.py before setup. After upgrading from a smaller scope set, re-run fitbit-mcp auth to re-authorise.

Configuration

Paths are overridable via environment variables:

Offline / cache-only mode

By default the server auto-syncs on demand, so query tools fetch fresh data without a cron job. Set FITBIT_MCP_OFFLINE=1 to run as a pure cache reader instead:

• No Fitbit credentials are required - the server never touches the token file.
• No live API calls are made. Auto-sync is disabled, and live=True, the live-only tools (fitbit_get_devices, fitbit_get_lifetime_stats, fitbit_get_goals), and fitbit_sync return a clear "offline mode" message instead of calling the API.
• Query tools serve whatever is in the local SQLite cache. Responses are tagged with "offline_mode": true.

Typical uses:

• Multi-device setups - one host runs fitbit-mcp sync (via cron/systemd) against a shared database; other hosts set FITBIT_MCP_OFFLINE=1 and point FITBIT_MCP_DB_PATH at the same cache, and only read. This keeps the Fitbit OAuth token (single-use, rotating) owned by exactly one host, avoiding refresh collisions.
• CI and privacy - run queries with no network access and no credentials.

Keeping the cache fresh is then the syncing host's job. Unset FITBIT_MCP_OFFLINE to return to on-demand auto-sync.

Rate limits

The Fitbit API allows 150 requests per hour. Activity and food log syncs sleep and retry automatically on a 429; the date-range data types instead mark that sync as rate_limited and pick up again on the next run. Be aware:

• Activity and food log syncs use 1 API call per day (no date-range endpoint available)
• A 30-day initial sync of either uses ~30 of your 150/hour quota
• Heart rate, sleep, weight, SpO2, HRV, AZM, breathing rate, skin temperature, core temperature, and cardio fitness use date-range endpoints and are much more efficient

Use live=False (the default) to query from cache and avoid API calls entirely.

Data safety

This project includes a pre-commit hook (scripts/check-no-data.sh) that prevents accidentally committing:

• Database files (*.db, *.db-journal, *.db-wal)
• Config/credentials (config/*.json)
• Large files (>100KB)

Install it after cloning:

ln -sf ../../scripts/check-no-data.sh .git/hooks/pre-commit

Importing existing data

If you have existing Fitbit data as JSON files (e.g. from a previous export or script), you can bulk-import them:

fitbit-mcp import --data-dir /path/to/json/files/

Expected file names: heart_rate.json, activity.json, exercises.json, sleep.json, weight.json, spo2.json, hrv.json. See src/fitbit_mcp/importer.py for the expected JSON format. Import currently covers these seven types only; the newer types (AZM, breathing rate, skin/core temperature, cardio fitness, food log) are populated via sync, not import.

Contributing

See CONTRIBUTING.md for development setup, the test workflow, and the pre-commit hook. Changes are tracked in CHANGELOG.md.

License

GPL-3.0-or-later