Wellness Project MCP
Hosted MCP server that syncs health data from Apple Health, Fitbit, Oura, and Google Health Connect, enabling Claude and ChatGPT to query workouts, sleep, nutrition, and recovery in plain English with interactive charts.
README
<p align="center"> <picture> <source media="(prefers-color-scheme: dark)" srcset="assets/lockup-dark.png"> <img src="assets/lockup-light.png" alt="Wellness Project" width="440"> </picture> </p>
Wellness Project MCP
Connect Apple Health, Fitbit, Oura, and Google Health Connect to Claude and ChatGPT. Wellness Project is a hosted Model Context Protocol (MCP) server: your devices sync into one account, and Claude or ChatGPT read your workouts, sleep, nutrition, and recovery in plain English. No exports, no copy-paste, no community server to self-host.
Wellness Project is a free AI health app. This repository documents its public MCP server: the tool catalog, the schemas, the inline-chart widgets, and how to connect any MCP client to it.
- App: https://wellnessproject.ai
- MCP endpoint:
https://wellnessproject.ai/api/mcp - 60 tools across workouts, nutrition, sleep, wearables, body metrics, labs, recovery, runs, and more
- 11 interactive chart widgets rendered inline in Claude and ChatGPT
Why this exists
Most health and fitness services ship no MCP server of their own. There is no official Fitbit MCP server, no Oura MCP server, no Hevy or TrainingPeaks MCP server, so Claude and ChatGPT cannot read that data on their own. The GitHub results for those queries are mostly unmaintained community repos you would have to run, refresh tokens for, and keep alive yourself.
Wellness Project is the hosted alternative. It connects to your devices over OAuth, ingests your workouts and recovery data, and exposes all of it to any MCP client through one authenticated endpoint. You connect once and it stays connected.
What you can ask
Once connected, you ask your AI naturally and it calls the matching tool:
- "How did my training look this week?"
- "Show my sleep for the last month."
- "How is my bench progressing?"
- "Log today's lunch: chicken burrito bowl, about 700 calories."
- "What was my resting heart rate trend vs my HRV this quarter?"
- "How many rest days did I take this month?"
Read tools pull your logged and synced data. Write tools log new entries. Chart tools render an interactive inline graph instead of a wall of numbers.
Connect Apple Health, Fitbit, or Oura to Claude
Connect your device once at wellnessproject.ai, then add the MCP server to Claude or ChatGPT. The same hosted endpoint serves both, so connecting Apple Health to Claude, Fitbit to ChatGPT, or Oura to either works the same way. The steps are below.
Connect to Claude
claude.ai (remote connector, recommended)
- Create a free account at https://wellnessproject.ai and connect a device (Settings → Integrations).
- In Claude, go to Settings → Connectors → Add custom connector.
- Paste
https://wellnessproject.ai/api/mcpand authorize with OAuth.
That is it. The connector carries across every conversation.
Desktop / scripted clients (API key)
Headless and desktop clients use a long-lived personal API key instead of the OAuth flow. Generate one in Settings → Claude inside the app. The server speaks streamable-HTTP MCP and authenticates with a bearer token:
POST https://wellnessproject.ai/api/mcp
Authorization: Bearer YOUR_PERSONAL_API_KEY
For a client that supports remote HTTP MCP servers with custom headers, point it at that URL with the Authorization header above. For a stdio-only client, bridge to it with mcp-remote:
{
"mcpServers": {
"wellness-project": {
"command": "npx",
"args": [
"mcp-remote",
"https://wellnessproject.ai/api/mcp",
"--header",
"Authorization: Bearer YOUR_PERSONAL_API_KEY"
]
}
}
}
Rotate or revoke the key from the same settings screen at any time.
Connect to ChatGPT
The same server backs ChatGPT. Connect Wellness Project as a custom connector / plugin pointed at https://wellnessproject.ai/api/mcp and authorize. The identical tools and widgets are available there.
Demo
A short screen recording of the Wellness Project connector running inside ChatGPT: connecting, asking questions in plain English, and the tools returning live data.
<video src="https://github.com/turnnoblindeye/wellness-project-mcp/raw/main/media/chatgpt-connector-demo.mp4" controls muted></video>
Inline chart widgets
Eleven tools render interactive charts inline through the MCP Apps UI extension, not plain text:
| Ask | Widget |
|---|---|
| "How did my week look?" | weekly Fit Score |
| "Break down my Fit Score" | today's Fit Score by component |
| "Show my workouts" | training activity + heart points |
| "Show that workout" | a single session's exercises and sets |
| "Show my macros" | calories + protein/carbs/fat vs targets |
| "How's my sleep?" | hours + sleep score |
| "Steps this week?" | step counts vs goal |
| "How's my running?" | running mileage |
| "My weight trend" | body weight + body-fat % |
| "Recovery trend" | resting heart rate + HRV |
| "How's my bench progressing?" | estimated 1RM progression |
Each chart has a 7d / 30d / 90d / 1y range toggle and hover tooltips, and reads live from your own account.
Tool catalog
The full list of 60 tools, grouped by domain, is in TOOLS.md. The machine-readable schemas (name, description, JSON Schema input, MCP annotations) are in catalog/tools.json and mirror exactly what the server returns from tools/list.
Every tool sets standard MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint). All tools have openWorldHint: false: they operate only on the authenticated user's own account data.
Data sources
| Source | Syncs |
|---|---|
| Apple Health (HealthKit) | steps, sleep, heart rate, HRV |
| Google Health Connect | steps, sleep, heart rate, HRV |
| Fitbit | steps, Active Zone Minutes, sleep stages, resting heart rate, HRV |
| Oura | sleep, readiness, heart rate, HRV |
| Manual / chat | workouts, meals, body metrics, labs, injuries, supplements, and more |
Privacy and auth
- Access is authenticated per user, by OAuth (claude.ai) or a personal API key (Desktop and scripted clients).
- API keys are stored hashed; the plaintext is shown to you once and never persisted.
- Every tool is scoped to the signed-in user. Row-level security enforces that a token can only read or write its own account.
- You can revoke a connector or rotate a key from app settings at any time.
Wellness Project is informational software, not a medical product.
FAQ
How do I connect Apple Health to Claude?
Create a free account at wellnessproject.ai and sync Apple Health (Settings, Integrations). In Claude, open Settings, Connectors, Add custom connector, paste https://wellnessproject.ai/api/mcp, and authorize. Claude can then read your Apple Health data in any conversation.
How do I connect Fitbit to Claude and ChatGPT?
Connect Fitbit once at wellnessproject.ai over OAuth. Then add https://wellnessproject.ai/api/mcp as a custom connector in Claude or as a custom connector in ChatGPT. One connection serves both assistants.
How do I connect Oura to Claude?
Connect Oura at wellnessproject.ai, then add the same MCP endpoint in Claude. Your Oura sleep, readiness, and HRV become available to ask about in plain English.
Is there an official Fitbit, Oura, or Apple Health MCP server?
No. None of these ship their own MCP server, and the community repos you find on GitHub are unofficial servers you have to run and refresh tokens for yourself. Wellness Project is the hosted alternative: connect your device once and it stays connected, with nothing to self-host.
Does this work with ChatGPT as well as Claude?
Yes. The same server backs both. Point a custom connector at https://wellnessproject.ai/api/mcp in either tool.
Do I have to self-host or run anything?
No. The server is hosted. You connect your devices through the app and add one URL to your AI client.
Is it free?
Yes. Wellness Project is free during early access.
Is my health data private?
Access is authenticated per user by OAuth or a personal API key. Every tool is scoped to the signed-in user, row-level security enforces that a token reads only its own account, and you can revoke access at any time.
Related
- Connect Apple Health to Claude: https://wellnessproject.ai/integrations/apple-health-to-claude
- Connect Fitbit to Claude and ChatGPT: https://wellnessproject.ai/integrations/fitbit-mcp
- Claude integration: https://wellnessproject.ai/integrations/claude
- ChatGPT integration: https://wellnessproject.ai/integrations/chatgpt
License
The tool catalog and schemas in this repository are published under CC BY 4.0 so MCP clients and directories can reference them freely. "Wellness Project" and the app itself remain the property of Wellness Project LLC.
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。