How jprbom/-health-dashboard-mcp works
Project type and purpose
• This is a Node.js “MCP server” (“Model Context Protocol”) that sits behind an AI assistant (e.g. Claude) to manage personal health and biometric data.
• Its purpose is to accept natural-language commands from the AI assistant, execute various “tools” (recording metrics, trend analysis, goal-setting, etc.), and return structured results.
• It also optionally integrates with external wearable APIs (Fitbit, Apple Watch, Garmin).
File structure
• Top-level:
• src/
• health_data/ (git-ignored local data store)
• .env (holds wearable API keys)
• package.json, .gitignore, README.md, LICENSE
• src/ contains:
• index.js – entrypoint that registers the MCP server
• tools/ – implementations of each “tool” (record_biometrics, get_health_summary, etc.)
• integrations/ – external API clients for Fitbit, Apple Watch, Garmin
• utils/ – shared helper functions (e.g., date parsing, file I/O)
• No frontend folder: the “client” is the AI assistant, communicating via an MCP wire protocol.
README for additional insights
• Lists available tools and their responsibilities.
• Specifies that all health data is stored locally under health_data/.
• Explains configuration for Claude Desktop to launch this server.
a. Main components
1. AI Assistant Client (Claude Desktop or other)
2. MCP Server (Node.js process)
• index.js (MCP server bootstrap)
• tools/ modules
• integrations/ modules
• utils/ modules
3. Local Data Store (health_data directory, file-based or lightweight DB)
4. External Wearable Services (Fitbit API, Apple Health, Garmin API)
b. Relationships and interactions
• AI Assistant → MCP Server: sends commands (JSON-RPC or similar)
• MCP Server → tools/: dispatches incoming commands to the correct tool
• tools/ → Local Data Store: read/write biometric data
• tools/ → integrations/: when syncing with wearable services, authenticate via .env keys and call external REST APIs
• MCP Server → AI Assistant: returns structured responses or insights
c. Key architectural patterns
• Modular “tool” pattern: each user-facing capability is encapsulated as its own module
• Adapter pattern for integrations: isolating wearable API clients behind a uniform interface
• Local persistence: simple file-based or embedded DB to keep data on the user’s machine
• Protocol adapter: the MCP server acts as a protocol translator between the AI assistant and the internal modules
d. Relevant technologies
• Node.js with npm modules
• JSON, REST for external service calls
• .env for configuration (dotenv)
• File I/O or lightweight DB (e.g., lowdb or similar) for health_data
Detailed breakdown
• A → B1 (MCP commands)
• B1 → B2 (dispatch to tools)
• B2 → C (read/write)
• B2 → B3 → D# (external sync)
• B2 → B1 → A (responses, insights)