Whoop

What whoop does

whoop exposes an authenticated WHOOP account to MCP-compatible assistants without providing write access to the account. Its standard tools cover profile and body measurements, recovery, sleep, workouts, and physiological cycles. Records can be retrieved by date range or by individual ID, making it suitable for both broad reviews and targeted lookups.

The server also turns recorded data into higher-level views. It can produce a daily snapshot, weekly summary, calendar-style date grids, personal metric baselines, sleep-debt analysis, period comparisons, and trend analysis. Trend results include direction, variability, and anomaly detection based on linear regression. Specialized HYROX and triathlon tracking tools are also included in the broader distribution.

How it works

After setup, the server runs on the local machine and communicates with the selected assistant over MCP. On its first start, it opens WHOOP in a browser so the user can approve access through the configured developer application. The assistant receives the resulting tool output and can use it to answer health questions or build summaries.

Access is read-only: the server does not create, edit, or delete WHOOP account data. A local aggregate mode can restrict access to five summary tools and hide per-record data, although this reduces exposure rather than anonymizing the information. Health results are sent to the AI provider used for the conversation, just like other content shared in that chat. Optional telemetry is disabled unless explicitly enabled and excludes health data and chat text.

Setup and configuration

Install Node.js 20 or later, maintain an active WHOOP membership, and create a WHOOP Developer App with this redirect URL:

http://localhost:3000/callback

The npm package can configure supported clients with commands such as:

npx -y whoop-ai-mcp@latest setup --client=claude-desktop

The setup command prompts for the developer app's client ID and secret, then prints client-specific registration or configuration instructions where needed. Supported setup targets include Claude Desktop, Claude Code, Codex, Gemini CLI, Cursor, VS Code, GitHub Copilot, OpenClaw, and Perplexity for macOS. After applying the configuration, fully restart the assistant and authorize WHOOP when the server starts.

Remote clients such as Claude web/mobile and ChatGPT require a public HTTPS /mcp endpoint, host-side WHOOP authorization, and compatible connector OAuth. The setup commands for those targets print deployment guidance rather than hosting the service or collecting remote credentials.

Tools and capabilities

The core read-only tools provide:

• Profile and body measurements, including name, email, height, weight, and maximum heart rate.
• Date-range recovery data with HRV, resting heart rate, SpO2, and skin temperature.
• Sleep data with stages, duration, respiratory rate, and performance scores.
• Workout data with strain, heart-rate zones, calories, and sport type.
• Cycle data with strain, calories, and heart-rate information.
• Individual sleep, workout, and cycle lookups by record ID.
• Daily, weekly, and calendar-based health summaries.
• Personal baselines, sleep-debt analysis, period comparisons, and metric trends.

The get_today tool combines recovery, the latest sleep, current strain, and the latest workout in one response. Additional HYROX and triathlon functionality supports session reconstruction, training-load views, readiness evidence, audit operations, and a local performance dashboard.

Limitations and notes

The health statistics describe recorded WHOOP history; they are not medical diagnoses or predictions. Results depend on the data available in the authenticated account, so summaries may contain gaps when records are missing. Aggregate mode limits the available tools but is not anonymization. Support for resources, prompts, remote connections, and some client-specific features varies by assistant. A native Rust implementation is available as an alternative for users who do not want a Node.js runtime.