Heavenly Health Protocol
Enables AI clients to securely access, query, and mutate normalized user-controlled health data (e.g., from Apple Health or Supabase) through a bounded set of MCP tools, with optional OAuth and sandboxed deployment.
README
Heavenly Health Protocol
Connect your health data — Apple Watch, Fitbit, Garmin, and more — to your AI agent. Privately, on your own computer, with your own storage.
Your agent (Claude, ChatGPT, Codex, Hermes, OpenClaw, …) gets bounded access to exactly the metrics you allow. Your credentials and raw health records never leave your control.
Get started
Requirements: Python 3.10+ and uv.
git clone https://github.com/obirimensah05/heavenly-health-protocol.git
cd heavenly-health-protocol
uv sync
uv tool install .
heavenly setup
heavenly setup walks you through everything. There is nothing to configure
first, no Docker, and no jargon.
The onboarding questions
Setup is a short guided conversation. These are the questions, in order:
- Which fitness or health device do you use? Apple Watch / iPhone · Fitbit / Pixel Watch · Garmin · WHOOP · Oura Ring · other Android wearables
- Which app is your health data's source of truth? Apple Health · Google Health API · Garmin Connect · WHOOP · Oura · Health Connect
- Where should your agent-readable health data live? Your own free Supabase project works today; Obsidian, local SQLite, Google Drive, and iCloud Drive are on the roadmap.
- Which AI agent should read it — and where does it run? Claude Code, Claude, ChatGPT, Codex, Hermes, OpenClaw, Perplexity, or any other compatible agent, on this computer or in the cloud.
- When should your health analysis arrive? Daily, every 3 days, or weekly · morning or evening · your timezone is detected automatically.
- Which metrics may it track? A minimal allowlist you approve. Clinical records, medication, reproductive data, ECG, and location stay off by default.
- Start and connect. Heavenly starts on your computer and shows the one-line connection step for your agent.
That's it. Advanced extras — the Docker runtime, remote access for cloud agents, the agent sandbox — are offered at the very end, default to off, and most people never need them.
What works today
| Your setup | Status |
|---|---|
| Apple Watch / iPhone via Apple Health + Health Auto Export | ✅ Works today |
| Fitbit / Pixel Watch via the Google Health API v4 | ✅ Works today |
| WHOOP | ✅ Works today — data needs an active WHOOP membership |
| Oura | ✅ Works today |
| Garmin via Garmin Connect | ✅ Built — requires Garmin Developer Program approval |
| Data you already keep in Supabase | ✅ Works today |
| Android Health Connect | 🔜 Planned (spec ready) |
Choosing a planned source still completes onboarding — your choice is recorded and nothing has to be redone when its adapter ships.
Signing in to WHOOP or Oura
Both take about two minutes and follow the same pattern:
- Create a (free) developer app on the provider's site — WHOOP or Oura — and note its client ID, client secret, and the redirect URL you register.
- Put those four lines in an owner-only env file
(
~/.config/heavenly/whoop.envoroura.env— exact keys in docs/providers). - Sign in with your normal account:
heavenly provider whoop import-client ~/.config/heavenly/whoop.env
heavenly provider whoop connect # browser opens; paste the redirected URL back
heavenly provider whoop sync
(Swap whoop for oura for an Oura Ring.) Your sign-in tokens are stored in
the operating-system credential vault — never in files, your notes, or Git.
How it stays private
- Runs on your computer and listens only on localhost by default.
- Health data lives in storage you own; keys stay in an owner-only local file and the operating-system credential vault.
- Your agent sees a bounded, metric-allowlisted read surface — never raw provider payloads, credentials, or arbitrary SQL.
- Any write to your health data needs your explicit approval in the terminal. An agent can never approve its own writes.
Details: SECURITY.md and the security design.
For technical operators (optional)
Everything below is opt-in and never required for the normal path above:
- Remote access for cloud agents through Cloudflare Managed OAuth — Deployment
- Hardened Docker runtime — Deployment
- Generic CLI-agent sandbox — Agent sandbox
- Architecture and MCP tool policy — Architecture, MCP tool policy
- Provider onboarding contracts — Providers
Development
uv sync --extra dev
uv run ruff check src tests
uv run pyright src
uv run --extra dev pytest
The release process exports only a validated tracked-file manifest into a fresh
Git history. Local .env, credential, log, and state files are excluded and
rejected by the public-release guard.
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.