quantified-self-mcp

# Quantified Self MCP Simple SQL-based quantified self tracking system with prompt-driven AI tools. ## What This Does - **Captures quantified self data** (workouts, food, sleep) from photos/text - **AI processes inputs** through MCP tools that work with SQL database - **Evolves schema naturally** - AI adds columns as needed - **Generates insights** through SQL analysis ## Architecture ### MCP Server (AI Interface) 5 tools that let AI work with SQL database: - `list_tables` - discover existing data - `create_table` - make new tables - `add_column` - evolve schemas - `insert_data` - store extracted data - `query_data` - analyze patterns ### n8n Automation (Processing Pipeline) 3 agents that handle the workflow: - **Input Agent** - extracts data from photos/text - **Extraction Agent** - uses MCP tools to store data - **Analysis Agent** - generates insights with SQL ### Database (Supabase PostgreSQL) Simple tables with consistent patterns: ```sql CREATE TABLE workouts ( id UUID PRIMARY KEY, date TIMESTAMP NOT NULL, exercise TEXT NOT NULL, sets INTEGER, reps INTEGER, weight REAL, created_at TIMESTAMP DEFAULT NOW() ); ``` ## Quick Start 1. **Setup Database**: Follow `docs/supabase-setup.md` 2. **Build MCP Server**: See `docs/mcp-server-implementation.md` 3. **Test Integration**: Use `docs/implementation-guide.md` ## Example Workflow **Input**: Photo of CrossFit whiteboard "21-15-9 Thrusters (95lbs), Pull-ups" **Processing**: 1. Input Agent extracts: workout type, exercises, rep scheme, weights 2. Extraction Agent uses `list_tables()` → sees workouts table 3. Extraction Agent uses `insert_data()` → stores 6 rows (3 sets × 2 exercises) 4. Analysis Agent uses `query_data()` → finds patterns and progress **Output**: Structured workout data + insights about progress ## Project Structure ``` spreadsheet-mcp/ ├── README.md ├── docs/ │ ├── README.md # Architecture overview │ ├── supabase-setup.md # Database setup │ ├── mcp-server-implementation.md # MCP server code │ ├── mcp-tool-examples.md # Tool usage patterns │ ├── n8n-agent-prompts.md # Agent prompt templates │ └── implementation-guide.md # Step-by-step guide └── sql/ └── sample_tables.sql # Database schema + sample data ``` ## Key Features ### Prompt-Driven AI Each MCP tool has rich prompt context that teaches the AI: - When to create new tables vs extend existing ones - How to handle schema evolution gracefully - What constitutes good vs bad data modeling decisions ### Natural Schema Evolution ```python # User: "I want to track RPE now" # AI automatically: list_tables() # → sees workouts table add_column("workouts", {"name": "rpe", "type": "INTEGER"}) # → adds RPE column insert_data("workouts", {..., "rpe": 8}) # → stores workout with RPE ``` ### Cross-Domain Analysis ```sql -- Find workout performance on high-fiber days SELECT w.exercise, w.weight, f.fiber FROM workouts w JOIN food f ON DATE(w.date) = DATE(f.date) WHERE f.fiber > 10; ``` ## Ready for Hackathon Everything your engineer needs is in the `docs/` folder: - Complete MCP server implementation - Supabase setup with metadata system - n8n agent prompts for automation - Step-by-step implementation guide **Time Estimate**: - Core MCP server: 2-3 hours - Database setup: 30 minutes - Basic testing: 1 hour - n8n integration: 4-6 hours The system is designed to be simple but powerful - normal SQL tables with AI tools that know how to use them intelligently.