withings-mcp

MCP server for the Withings Health API with OAuth, local SQLite cache, and trend analysis.

What makes this different from other Withings MCP servers:

• Local SQLite cache for fast offline queries and historical trend analysis

• Incremental sync - only fetches new data since last sync

• Broad Withings coverage: 17 body-composition metrics plus sleep, daily activity, workouts, and ECG/AFib

• Automatic OAuth token refresh (access tokens: 3h, refresh tokens: 1 year)

• Zero dependencies beyond mcp (HTTP via stdlib)

• Python 3.13+

Tools

The cache-backed query tools auto-sync when their data is stale, and accept live=True to bypass the cache and fetch straight from the Withings API. withings_get_heart and withings_get_devices are always live. withings_get_sleep(detail=True) returns minute-by-minute sleep phases (live, up to 7 days per request).

Related MCP server: Withings MCP Server

Prerequisites

• Python 3.13+

• uv (recommended) or pip

• A Withings developer account and registered application

Installation

pip install withings-mcp

Or run it without installing with uvx withings-mcp. For development from a clone:

git clone https://github.com/partymola/withings-mcp.git
cd withings-mcp
uv venv --python 3.13 .venv
uv pip install -e .

Setup

1. Register a Withings app

1. Go to https://developer.withings.com/dashboard

2. Create a new application

3. Set the callback URL to http://localhost:8585

4. Note your Client ID and Client Secret

2. Authenticate

.venv/bin/withings-mcp auth

This opens your browser for Withings authorization. After approving, tokens are saved locally in config/.

3. Register with Claude Code

claude mcp add -s user withings -- /path/to/withings-mcp/.venv/bin/withings-mcp

4. First sync

In Claude Code, say: "Sync my Withings data"

This runs withings_sync to populate the local cache. Subsequent syncs only fetch new data.

If anything does not work - no data where you expect it, a sync that stops happening - run withings-mcp doctor. It reports the paths and credentials actually in use and what needs fixing, without making an API call.

You can also sync from the command line without an MCP client:

.venv/bin/withings-mcp sync                      # all data types, last 30 days
.venv/bin/withings-mcp sync --types body,sleep   # a subset
.venv/bin/withings-mcp sync --days 90            # deeper history on first sync

CLI

withings-mcp              Start the MCP server (stdio transport)
withings-mcp auth         Interactive OAuth setup (opens the browser)
withings-mcp sync         Sync data to the local cache (--types, --days)
withings-mcp doctor       Check the setup and report what needs fixing
withings-mcp --version    Print the installed package version

Configuration

Example Prompts

• "Sync my Withings data"

• "Show my weight for the last 3 months"

• "How has my sleep changed this year?"

• "Compare my body composition this month vs last month"

• "What workouts did I do in March?"

• "What Withings devices do I have connected?"

• "Show my sleep trends quarterly"

Development

# Install with dev dependencies
uv pip install -e . && uv pip install pytest

# Run tests (all use in-memory SQLite with fictional data)
.venv/bin/python -m pytest tests/ -v

Security

• Read-only: No tools modify data on Withings servers

• Local storage: Health data stays in your local SQLite database

• Token storage: OAuth tokens stored in config/ (gitignored, file permissions 0600)

• Error messages: Never contain health data values - only status codes

• Pre-commit hook: An optional hook (scripts/check-no-data.sh) blocks database files and credentials from commits - install it with the one-liner in CONTRIBUTING.md

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