WHOOP ChatGPT App

A ChatGPT Apps SDK project that connects WHOOP health tracker data to ChatGPT through a remote MCP server and renders an interactive WHOOP dashboard widget in ChatGPT.

The app is a read-only health-data connector. It can search WHOOP records, fetch individual records, summarize recent recovery/sleep/strain/workout data, and render a dashboard UI from either freshly loaded WHOOP data or previously returned structured content.

This project is for software integration and personal health-data exploration. It is not medical advice.

Screenshots

These screenshots use synthetic demo data and are safe to publish.

Desktop	Mobile

Features

• ChatGPT Apps SDK-compatible MCP server at /mcp
• Connector-style search and fetch tools for read-only WHOOP data retrieval
• WHOOP OAuth helper for local development
• React widget rendered inside ChatGPT
• Dashboard cards for recovery, strain, sleep performance, and latest workout
• Recent records timeline with filters for recovery, sleep, workout, and cycle strain
• Widget actions for refresh, interpretation follow-up, and fullscreen display
• Synthetic demo mode for safe screenshots and README assets

Architecture

ChatGPT
  |
  | remote MCP over HTTPS
  v
server/
  - MCP tools
  - WHOOP OAuth helper
  - WHOOP API client
  - widget resource registration
  |
  | embeds bundled widget resource
  v
web/
  - React dashboard widget
  - MCP Apps bridge integration

App shape:

• Archetype: react-widget
• Architecture: decoupled data/render
• Connector surface: standard read-only search and fetch tools
• Widget rendering: render_whoop_dashboard

Tools

All tools are read-only.

Tool	Purpose
search(query)	Finds recent WHOOP profile, recovery, cycle, sleep, and workout records.
fetch(id)	Fetches one record returned by search.
get_whoop_summary(days, includeProfile)	Returns a compact model-readable WHOOP summary.
render_whoop_dashboard(days, includeProfile, dashboard)	Renders the React widget. It can load data directly using days/includeProfile, or render a dashboard object returned by get_whoop_summary.

Requirements

• Node.js 20 or newer
• npm
• A WHOOP Developer app for live data
• An HTTPS tunnel for ChatGPT Developer Mode testing, such as ngrok

Quick Start

npm install
cp .env.example .env
npm run build
npm start

The server listens on:

http://localhost:8787/mcp

WHOOP Setup

Create a WHOOP Developer app and configure these read scopes:

read:profile read:body_measurement read:cycles read:recovery read:sleep read:workout

For local development:

1. Start a public HTTPS tunnel to port 8787.
2. Set PUBLIC_BASE_URL to the tunnel origin.
3. Set WHOOP_CLIENT_ID, WHOOP_CLIENT_SECRET, and WHOOP_REDIRECT_URI in .env.
4. Register the same redirect URI in WHOOP, for example https://example.ngrok.app/oauth/callback.
5. Restart the server.
6. Open https://example.ngrok.app/oauth/start in your browser and link WHOOP.

The local OAuth helper stores the WHOOP access token in server memory. Restarting the server clears the link.

ChatGPT Developer Mode

ChatGPT requires a public HTTPS MCP URL for remote app testing.

npm start
ngrok http 8787

Use the tunneled MCP URL:

https://example.ngrok.app/mcp

In ChatGPT:

1. Enable Developer Mode under Settings -> Apps & Connectors -> Advanced settings.
2. Create a new app/connector.
3. Paste the tunneled /mcp URL.
4. Use No Auth for local development with CHATGPT_APP_AUTH=none.
5. Refresh or reconnect the app after tool, metadata, or widget changes.

Example prompt:

Use the Whoop app to show my WHOOP dashboard for the last 7 days.

Environment Variables

Copy .env.example to .env and fill in local values.

Variable	Required	Description
PORT	No	Local HTTP port. Defaults to 8787.
PUBLIC_BASE_URL	Yes for OAuth/ChatGPT	Public origin for OAuth redirects and metadata.
WHOOP_CLIENT_ID	Yes for OAuth	WHOOP Developer app client ID.
WHOOP_CLIENT_SECRET	Yes for OAuth	WHOOP Developer app client secret. Do not commit it.
WHOOP_REDIRECT_URI	Yes for OAuth	Redirect URI registered in WHOOP.
CHATGPT_APP_AUTH	No	Use none for local Developer Mode testing.
WHOOP_ACCESS_TOKEN	No	Local single-user shortcut. Prefer OAuth for normal testing.
AUTHORIZATION_SERVER_URL	Production only	OAuth 2.1 authorization server for production ChatGPT linking.
WHOOP_API_BASE_URL	No	WHOOP API override for tests or proxies.

Development

npm run dev      # run the TypeScript server with tsx
npm run build    # build web widget and server
npm run check    # type-check web and server
npm start        # run the built server

MCP Inspector:

npx @modelcontextprotocol/inspector@latest --server-url http://localhost:8787/mcp --transport http

Project Structure

.
├── docs/screenshots/        # publishable screenshots with synthetic data
├── server/
│   └── src/
│       ├── index.ts         # MCP server, tools, OAuth routes, resource registration
│       ├── whoop-client.ts  # WHOOP API client
│       └── whoop-format.ts  # record formatting and dashboard summarization
├── web/
│   └── src/
│       ├── main.tsx         # React widget and Apps bridge integration
│       └── styles.css       # responsive widget styling
├── .env.example
├── package.json
└── README.md

Security and Privacy

• .env is ignored by git.
• Do not commit WHOOP client secrets, access tokens, refresh tokens, or tunnel URLs.
• WHOOP tokens stay server-side and are never exposed to the widget.
• The local OAuth helper stores tokens in memory only.
• Screenshots in docs/screenshots use synthetic data.
• Production apps should validate token issuer, audience/resource, expiry, and scopes before calling WHOOP.

Production Notes

The local setup uses CHATGPT_APP_AUTH=none so ChatGPT Developer Mode can connect while the server uses the locally linked WHOOP token. For production, put a ChatGPT-compatible OAuth 2.1 authorization layer in front of WHOOP and advertise OAuth security schemes from the MCP server.

Production hosting should include:

• Stable public HTTPS endpoint
• Secure secret storage
• Token verification and refresh handling
• Request logging without sensitive data
• Rate-limit and WHOOP API error handling
• Re-test in ChatGPT Developer Mode before submission

Open Source Checklist

Before publishing:

• Confirm .env is not staged.
• Replace any local tunnel URLs in docs or examples with placeholders.
• Run npm run check.
• Run npm run build.
• Review screenshots for synthetic data only.

License

MIT. See LICENSE.