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.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。