WHOOP MCP Unofficial bridge for AI agents

whoopmcp.vercel.app · open source · unofficial

Give your AIa body signal.

WHOOP MCP is a local-first bridge that lets MCP-aware assistants understand your WHOOP recovery, sleep, strain and workouts. Not another dashboard: a safer way for your agent to reason about training, recovery, focus and sleep.

Install in 3 commands Give this to your agent
npx -y whoop-mcp-unofficial setup
  • 20 toolsagent-discoverable WHOOP actions
  • 6 resourceslatest signal and summaries
  • 0 write scopesread-only WHOOP OAuth
  • Local tokenskept under ~/.whoop-mcp
Unofficial. This project is not affiliated with, endorsed by, or sponsored by WHOOP, Inc. Use it only with your own WHOOP account and the official WHOOP OAuth flow.
Primary sitewhoopmcp.vercel.app SourceGitHub repo Installnpm package MirrorGitHub Pages

The product idea

From wearable data to a useful next move.

Most health data stops at a chart. This bridge gives your assistant enough context to help you make a cleaner decision today.

01

Read the signal

Recovery, sleep, strain, workouts and profile data come from your authorized WHOOP account.

02

Compress the noise

The MCP server turns raw endpoints into concise summaries your assistant can reason about.

03

Choose the action

Your AI can suggest whether to push, maintain, deload, protect sleep or investigate a pattern.

Data boundary

Clear about what WHOOP exposes.

The bridge uses the official WHOOP OAuth API. It can expose the full API payload when you opt in, but it does not claim access to continuous sensor streams.

Yes

Processed WHOOP records

Recovery, HRV, resting heart rate, sleep stages, cycles, strain, workouts, heart-rate zones, profile and body measurements.

Raw

Raw API payloads

privacy_mode=raw returns the upstream JSON from supported WHOOP endpoints. It is opt-in because health data is sensitive.

No

No sensor streams

Continuous or second-by-second heart-rate samples are not available through the official WHOOP API, so they are outside this MCP today.

Local-first architecture

Designed for trust before scale.

Your agent gets useful context, not your secrets. WHOOP OAuth stays inside the official API boundary, tokens stay local, and the MCP exposes clear tools/resources instead of vague scraped data.

Official API boundary No private WHOOP endpoints or misleading raw sensor claims.
Agent-readable contracts whoop_capabilities explains tools, limits and privacy modes.
Action-oriented summaries Daily and weekly workflows turn records into practical next moves.
Architecture diagram showing WHOOP OAuth flowing through a local MCP server into AI clients

Questions people ask first

Precise answers before anyone installs.

A good health-data bridge should be explicit about what it can read, what it cannot read, and where the trust boundary sits.

Does this fetch raw sensor data?

No. raw means the full JSON returned by supported WHOOP API endpoints. It does not mean continuous heart-rate samples, accelerometer data or raw device telemetry.

What should my agent call first?

Start with whoop_capabilities, then whoop_connection_status, then whoop_daily_summary or whoop_weekly_summary.

Where do secrets live?

WHOOP credentials and tokens stay on your machine under ~/.whoop-mcp/. Tools intentionally do not return OAuth token values.

Is this medical advice?

No. It gives recovery, sleep, training and performance context for your own agent workflows. It is not diagnosis, treatment or emergency monitoring.

Read the full FAQ See the roadmap

Guided setup

Install it without becoming a developer.

The workflow is intentionally calm: create a WHOOP app, run three commands, add one MCP server to your AI client.

Path A

I am a human

Create a WHOOP Developer app, paste the credentials into setup, authorize once in the browser, then use the MCP config.

Path B

I want my agent to help

Copy the delegation prompt below. It tells your assistant exactly what to install, what not to expose, and how to verify safely.

Jump to agent prompt
Step 1

Create your WHOOP developer app

Open the WHOOP Developer Dashboard, create an app, and use the exact values below.

Redirect URL http://127.0.0.1:3000/callback
Read scopes read:recovery read:cycles read:workout read:sleep read:profile read:body_measurement

After saving, WHOOP gives you a Client ID and Client Secret. You will paste them into the local installer.

Open WHOOP Developer Dashboard
Step 2

Authorize locally

The CLI collects your app credentials, opens WHOOP in your browser, then verifies that the bridge is healthy.

npx -y whoop-mcp-unofficial setup
npx -y whoop-mcp-unofficial auth
npx -y whoop-mcp-unofficial doctor

The authorization callback happens on 127.0.0.1. Your tokens are written locally, not pasted into chat.

Step 3

Connect your AI assistant

Use it with Claude Desktop, Cursor, Windsurf, Hermes, OpenClaw or any client that supports MCP over stdio.

Claude Cursor Windsurf Hermes OpenClaw Generic MCP
Show the MCP config 
{
  "mcpServers": {
    "whoop": {
      "command": "npx",
      "args": ["-y", "whoop-mcp-unofficial"]
    }
  }
}

Prompts that feel like a coach

Ask for decisions, not data dumps.

These prompts help your assistant turn WHOOP context into a concrete training, recovery or focus decision.

Morning

Design today

Use my latest WHOOP recovery, sleep and strain. Tell me if today should be a push, maintain or recovery day, then give me one action that matters most tonight.

Training

Choose intensity

Review my last 3 days of workouts and recovery. Recommend today's training intensity and explain the tradeoff in plain English.

Weekly

Find the bottleneck

Review my last 7 days of WHOOP data. Identify the biggest bottleneck, the best signal, and a simple plan for next week.

Diagnostic

Investigate a drop

Look at the last 14 days. If my recovery dropped, find the strongest likely signal behind it and show the numbers without giving medical advice.

Trust boundary

Built for sensitive health data.

WHOOP data is personal. The bridge is designed so assistants get useful health context without needing to see OAuth tokens or write back to your account.

Read-only

Only WHOOP read scopes are requested. The bridge does not create, edit or delete WHOOP data.

Local tokens

Token files live under ~/.whoop-mcp/ with user-only permissions.

Structured by default

Summaries are the main path. Raw payloads should be explicit, not accidental.

No medical advice

This is a developer tool for reflection and coaching context, not diagnosis or treatment.

For AI agents

A delegation prompt that does not guess.

If a non-technical user wants their assistant to install the bridge, this prompt gives the agent enough structure to proceed safely.

Install the unofficial WHOOP MCP server for me.
Repository: https://github.com/davidmosiah/whoop-mcp.

Help me create a WHOOP developer app with:
- Redirect URI: http://127.0.0.1:3000/callback
- Read-only WHOOP scopes

Run setup, auth and doctor with npx.
Add the MCP config to my client.
Verify the connection.
Keep tokens local and never print secrets.

For builders

Open source, with a clear edge.

Built by David Mosiah for people exploring practical AI-agent systems around personal health, recovery and performance. The project is small enough to understand, but serious about privacy, OAuth and agent ergonomics.

Open an issue Contribute Roadmap

Open source health agents

Make AI assistants better at respecting the body.

Star the repo, inspect the implementation, open issues or adapt the bridge for your own agent workflow.

Open GitHub npm package MCP Registry