whoopmcp
A read-only MCP server for the WHOOP API v2 that lets you query and analyze your own recovery, sleep, strain, cycles, and workout data. Note: currently a pre-alpha scaffold with stubbed internals.
README
whoopmcp
A read-only MCP server for the WHOOP API v2. It lets an MCP client — Claude Desktop, Claude Code, Cursor, or anything else that speaks the protocol — read and analyse your own WHOOP data: recovery, sleep, strain, cycles and workouts.
Runs locally. Your WHOOP credentials never leave your machine.
Status: pre-alpha scaffold. The structure, tool surface, configuration and test harness are in place; the network and analysis internals are stubbed and raise
NotImplementedError. See Roadmap. Nothing here talks to WHOOP yet.
Not affiliated with WHOOP, Inc. "WHOOP" is their trademark. This is an independent client of their public developer API.
What it does
| Area | Tools |
|---|---|
| Auth | whoop_auth_status, whoop_login, whoop_complete_login, whoop_logout |
| Profile | get_profile, get_body_measurement |
| Records | list_recoveries, list_sleeps, list_cycles, list_workouts, get_sleep, get_workout |
| Analysis | summarize_period, metric_trend, correlate_metrics, compare_periods |
Every data tool is annotated readOnlyHint. There is no write path to your
WHOOP account in this server — the one mutating endpoint WHOOP exposes
(DELETE /v2/user/access) is deliberately not wired up, so a model cannot
revoke your grant. whoop_logout only deletes the token stored on your own
disk.
Questions it is meant to answer:
- "How did my recovery trend over the last month?"
- "Is my HRV correlated with how long I sleep?"
- "Compare my strain in July against June."
What it is not
It reports numbers and the sample size behind them. It is not a medical device, it does not diagnose, and a correlation across a few weeks of your own data is not a causal finding. Talk to a clinician about health decisions.
Install
Requires Python 3.13 or 3.14 — the two newest stable releases, which are the two CI tests.
uvx whoopmcp # run without installing
# or
pip install whoopmcp
Setup
You need your own WHOOP developer app — this server ships no shared credentials, by design. Full walkthrough in docs/SETUP.md. The short version:
- Create an app at developer.whoop.com and note the client ID and secret.
- Register a redirect URL. WHOOP does not accept
http://, includinghttp://localhost— usehttps://or a custom scheme such aswhoopmcp://callback. - Point the server at them via environment variables.
Claude Desktop / Claude Code
{
"mcpServers": {
"whoop": {
"command": "uvx",
"args": ["whoopmcp"],
"env": {
"WHOOP_CLIENT_ID": "your-client-id",
"WHOOP_CLIENT_SECRET": "your-client-secret",
"WHOOP_REDIRECT_URI": "whoopmcp://callback"
}
}
}
}
Then ask your client to run whoop_login, open the URL it returns, approve
the consent screen, and pass the code and state from the redirect back
via whoop_complete_login.
Configuration
| Variable | Required | Default | Meaning |
|---|---|---|---|
WHOOP_CLIENT_ID |
yes | — | OAuth client ID from the WHOOP dashboard |
WHOOP_CLIENT_SECRET |
yes | — | OAuth client secret |
WHOOP_REDIRECT_URI |
yes | — | Must match a registered redirect URL exactly |
WHOOPMCP_SCOPES |
no | all read scopes + offline |
Space-separated scope list |
WHOOPMCP_TOKEN_BACKEND |
no | file |
file or keyring |
WHOOPMCP_STATE_DIR |
no | ~/.local/state/whoopmcp |
Token and cache location |
WHOOPMCP_CACHE |
no | false |
Cache responses on disk |
WHOOPMCP_TIMEOUT |
no | 30 |
Per-request timeout, seconds |
The offline scope is requested by default. Without it WHOOP issues no
refresh token and you would re-authorise through a browser every hour.
For a token in your OS keychain rather than a file on disk:
pip install 'whoopmcp[keyring]'
export WHOOPMCP_TOKEN_BACKEND=keyring
Recommended on Windows, where the default file backend cannot protect the
token: Windows uses ACLs rather than POSIX modes, so the 0600 the file
backend requests is ignored and the token lands world-readable. The server
warns when it first writes one.
Privacy
Read PRIVACY.md before connecting real data. The essential points:
- This server sends nothing to its maintainers. No telemetry, no
analytics, no phone-home. Traffic goes to
api.prod.whoop.comand nowhere else. - Your MCP client does send your data onward. Anything a tool returns is passed to whatever model your client is configured to use — Anthropic, OpenAI, a local model — under that provider's terms, not this project's. This is inherent to how MCP works, and it is health data. Know where it is going.
- Tokens are stored locally at mode
0600, or in your OS keychain. On Windows file modes are not enforced — use the keychain backend there. - Delete everything with
whoop_logout, then removeWHOOPMCP_STATE_DIR, then revoke the app in the WHOOP app under Settings.
Rate limits
WHOOP's documented defaults are 100 requests/minute and 10,000/day,
with X-RateLimit-* headers and a 429 on breach. Collections page at 25
records maximum. Ask for explicit date ranges; an unbounded walk over years
of history will exhaust the quota and the model's context window alike.
Confirmed with WHOOP: the limit is per application (your client_id),
shared across every member who has authorised it — not a separate budget
per member. Running this locally for one person, that distinction is
invisible. Hosting it for several, it is the whole budget: one member's
two-year backfill is roughly 110 requests, about a minute of the entire
app's per-minute quota (#9).
Development
git clone https://github.com/kurok/whoopmcp
cd whoopmcp
uv venv && uv pip install -e '.[dev,lint]'
pytest # tests
ruff check . && ruff format --check .
mypy # strict on src/
Built on the official Python SDK's MCPServer — the class FastMCP became
when the SDK went to 2.0. The layering is deliberate:
config.py environment -> Config, validated once at startup
auth.py OAuth 2.0 flow + token storage (file or keychain)
client.py one method per documented WHOOP endpoint, nothing more
analysis.py pure functions over already-fetched records
server.py MCP tool definitions; the only file that knows about MCP
analysis.py holds no network code and client.py holds no statistics, so
each can be tested without the other.
Roadmap
The scaffold is complete and CI is green; these are the tracked gaps.
| Issue | Work |
|---|---|
| #1 | OAuth token exchange and refresh |
| #2 | HTTP transport: bearer auth, 429 handling, pagination |
| #3 | Record shaping: metric extraction, summaries, trends, correlation |
| #4 | Auth tools |
| #5 | Data tools |
| #6 | Analysis tools |
Each stub raises NotImplementedError naming its issue, and the test suite
pins the contract each one must satisfy.
Contributing
See CONTRIBUTING.md. Bug reports and PRs welcome; by participating you agree to the Code of Conduct. To report a security issue, follow SECURITY.md rather than opening a public issue.
Prior art
Several other WHOOP MCP servers exist — among them AshwanthramKL/whoop-mcp, shashankswe2020-ux/whoop-mcp and JedPattersonn/whoop-mcp. If one of them already does what you need, use it.
License
MIT.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.