Health Hub MCP Server
Aggregates Android Health Connect data into daily summaries, time-series metrics, sleep details, and sync status, exposing them to LLM clients via MCP with caveats to prevent data fabrication.
README
Health Hub MCP Server
Personal MCP server that ingests Health Connect data from an Android companion app and exposes aggregated health metrics to LLM clients over the Model Context Protocol.
Android 14+ ──POST /ingest/health──────▶ ┌──────────────────┐
(companion app) │ │
Android 14+ ──POST /ingest/hc-webhook──▶ │ Health Hub │──MCP──▶ LLM
(HC Webhook client) │ SQLite │
browser ──GET /debug──▶ └──────────────────┘
The companion app reads raw records from Health Connect and pushes them to this server. The server normalizes, deduplicates, and rolls them up into daily aggregates, then exposes them to LLM clients (Claude, Perplexity, OpenCode) through five MCP tools. Raw events are never handed to the model by default — only rolled-up summaries with explicit caveats so the model cannot silently fabricate missing data.
Requirements
- Android 14+ with life-dashboard-companion-app v1.4.0+
- Any host with Docker and Docker Compose
- Domain with TLS (Caddy handles certificates automatically)
Quick Start
cp .env.example .env
# Edit .env: set DEVICE_TOKEN and MCP_TOKEN (run: openssl rand -hex 32)
docker compose up -d --build
curl -s http://localhost:8000/health
Companion Apps
The server accepts health data from two Android apps via separate endpoints. Both write to the same database and use content-hash deduplication — sending the same data from both apps will not double-count.
| App | Endpoint | Format | Best for |
|---|---|---|---|
| life-dashboard-companion-app v1.4.0 | POST /ingest/health |
source: "health_connect" required |
Incremental sync, 200 records per type |
| HC Webhook | POST /ingest/hc-webhook |
source optional, rmssd_millis HRV, extra types |
Full sync, all types in one POST |
Both endpoints require:
Authorization: Bearer <DEVICE_TOKEN>
Content-Type: application/json; charset=utf-8
The HC Webhook endpoint also accepts 5 additional Health Connect types not present in the companion app: skin_temperature, basal_metabolic_rate, vo2_max, intermenstrual_bleeding, body_water_mass. Unknown type keys are silently tolerated — only types in the server's registry are processed.
Connect Your Phone
See docs/ANDROID.md for step-by-step instructions.
Connect an LLM
Add to Claude Desktop config (claude_desktop_config.json):
{
"mcpServers": {
"health-hub": {
"url": "https://your-domain.example.com/mcp",
"headers": {"Authorization": "Bearer <YOUR_MCP_TOKEN>"}
}
}
}
For Perplexity, OpenCode, or other MCP clients, use the same URL and token. The Authorization header must include the Bearer prefix — the server rejects requests without it.
MCP Tools
| Tool | Description |
|---|---|
daily_summary |
Daily health rollups: steps, distance, calories, sleep, HR, weight |
metric_series |
Time series for any metric with day/week bucketing |
sleep_detail |
Detailed sleep sessions with stages (deep, REM, light, awake) |
get_sync_health |
Per-type sync status: last event, coverage, gaps, errors |
inspect_last_payload |
Raw payload metadata and app diagnostics |
Each tool returns an object (never a bare array) with a caveats: string[] field that warns the model about missing days, overlapping records, or aggregation changes. Responses are capped at 400 objects.
Known Limitations
- No record IDs from Health Connect — deduplication is content-hash based
- Multiple data sources (phone + watch) can cause double-counting for cumulative metrics. The server suppresses overlapping intervals, but this is a heuristic. Check the comparison panel on
/debug. - App sync cursor only moves forward — records added to Health Connect retroactively may not arrive
- Health Connect does not provide data older than 30 days without re-authorization — the server IS your archive. Back up
data/.
License
MIT — see LICENSE.
Companion App
life-dashboard-companion-app v1.4.0
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。