MCP Agent Tracker

# Enhanced Conversation Context Injection System ## Overview The Enhanced Conversation Context Injection System transforms your MCP Agent Tracker from a simple logging tool into an intelligent context-aware system that can inject relevant conversation history, user preferences, and project context into every interaction with your Cursor agent. ## Key Features ### 🌳 Decision Tree Context Analysis - **Hierarchical Topic Structure**: Organizes conversations into logical branches (technical, project, user, system) - **Active Branch Detection**: Identifies currently relevant conversation topics - **Context Relevance Scoring**: Prioritizes context based on recency, interaction type, and content length ### 🧠 Intelligent Context Generation - **Semantic Keyword Extraction**: Uses NLP techniques to identify meaningful terms - **Topic Categorization**: Automatically categorizes interactions into 10+ topic types - **User Preference Inference**: Learns user communication style and technical level - **Project Context Extraction**: Detects files, technologies, and project patterns ### 🚀 Context Injection for Cursor - **Enhanced Prompts**: Automatically injects relevant context into every prompt - **Personalized Responses**: Tailors responses based on user history and preferences - **Project Awareness**: Provides context about current project structure and tech stack - **Conversation Continuity**: Maintains context across multiple interactions ## Architecture ``` ┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐ │ Cursor Agent │───▶│ Context Injection│───▶│ Enhanced Prompt │ └─────────────────┘ │ Tool │ └─────────────────┘ └──────────────────┘ │ ▼ ┌──────────────────┐ │ Decision Tree │ │ Context Manager │ └──────────────────┘ │ ▼ ┌──────────────────┐ │ Database with │ │ Full Content │ │ Retention │ └──────────────────┘ ``` ## Installation & Setup ### 1. Database Migration First, run the database migration to add new context tracking fields: ```bash python migrate_database.py ``` This will: - Add new columns to existing tables - Create the `conversation_contexts` table - Add performance indexes - Support both PostgreSQL and SQLite ### 2. Environment Configuration Ensure your environment variables are set: ```bash # For PostgreSQL (recommended) DATABASE_URL=postgresql://mcp_user:mcp_password@localhost:5432/mcp_tracker # For SQLite (development) DB_PATH=./data/agent_tracker.db ``` ### 3. Start the Enhanced System The system will automatically start using the enhanced context tracking: ```bash # Start MCP server with context injection python main.py # Or use HTTP server python mcp_http_server.py ``` ## Usage ### Basic Context Injection The system automatically enhances every interaction with context. Simply use the existing MCP tools: ```python # Get conversation summary with enhanced context summary = get_conversation_summary() # Get detailed context analysis context = get_conversation_context() ``` ### Advanced Context Injection Use the new context injection tool to enhance prompts for Cursor: ```python # Inject context into a base prompt base_prompt = "Help me fix this bug in my React component" enhanced_prompt = inject_conversation_context(base_prompt) # The enhanced prompt will include: # - Recent conversation history # - User preferences and technical level # - Project context and tech stack # - Active conversation branches # - Semantic keywords and topics ``` ### HTTP API Usage ```bash # Inject context via HTTP curl -X POST "http://localhost:8000/tool/inject_conversation_context" \ -H "Content-Type: application/json" \ -d '{"prompt": "Help me optimize this database query"}' # Get detailed context analysis curl -X POST "http://localhost:8000/tool/get_conversation_context" \ -H "Content-Type: application/json" \ -d '{}' ``` ## Context Injection Examples ### Example 1: Code Review Request **Base Prompt:** ``` "Can you review this Python function for best practices?" ``` **Enhanced Prompt:** ``` === CONVERSATION CONTEXT (Decision Tree Enhanced) === 🌳 ACTIVE CONVERSATION BRANCHES: • coding: Working on Python backend API with FastAPI • testing: Implementing unit tests for database models 📋 SUMMARY: Active conversation branches: coding (3 interactions), testing (2 interactions) 🎯 KEY TOPICS: coding, testing, database, api 👤 USER PREFERENCES: Technical Level: intermediate; Preferred Topics: coding, testing; Communication Style: collaborative 📁 PROJECT CONTEXT: Files: main.py, models.py, test_models.py; Tech Stack: backend: fastapi, database: postgresql 🕒 RECENT INTERACTIONS: 🔥 conversation_turn: Implementing user authentication system with JWT tokens... ⚡ tool_call: Running database migration for user table... 🔍 FREQUENT KEYWORDS: python, fastapi, database, test, api, user, authentication === END CONTEXT === Can you review this Python function for best practices? ``` ### Example 2: Debugging Request **Base Prompt:** ``` "Why is my Docker container failing to start?" ``` **Enhanced Prompt:** ``` === CONVERSATION CONTEXT (Decision Tree Enhanced) === 🌳 ACTIVE CONVERSATION BRANCHES: • deployment: Setting up Docker containers for microservices • debugging: Troubleshooting container startup issues 📋 SUMMARY: Active conversation branches: deployment (5 interactions), debugging (3 interactions) 🎯 KEY TOPICS: deployment, debugging, docker, microservices 👤 USER PREFERENCES: Technical Level: advanced; Preferred Topics: deployment, debugging; Communication Style: direct 📁 PROJECT CONTEXT: Files: docker-compose.yml, Dockerfile, .env; Tech Stack: devops: docker, kubernetes; backend: node, python 🕒 RECENT INTERACTIONS: 🔥 conversation_turn: Setting up multi-service Docker environment with shared networks... ⚡ tool_call: Checking container logs and resource usage... 🔍 FREQUENT KEYWORDS: docker, container, service, network, environment, compose, kubernetes === END CONTEXT === Why is my Docker container failing to start? ``` ## Context Types & Categories ### Topic Categories - **coding**: Programming, algorithms, data structures - **architecture**: System design, patterns, scalability - **deployment**: CI/CD, containers, infrastructure - **testing**: Unit tests, integration tests, QA - **debugging**: Error handling, troubleshooting, monitoring - **project_management**: Planning, collaboration, timelines - **data_analysis**: Analytics, visualization, ML/AI - **user_experience**: UI/UX, design, usability - **system_administration**: Servers, security, maintenance - **documentation**: Guides, APIs, knowledge bases ### User Preference Types - **Technical Level**: beginner, intermediate, advanced - **Communication Style**: direct, collaborative, analytical - **Preferred Topics**: Most discussed technical areas - **Project Focus**: Current project emphasis ### Project Context Types - **Files**: Source code, configuration files - **Technology Stack**: Frontend, backend, database, DevOps tools - **Project Names**: Repository names, domain references - **Architecture Patterns**: Microservices, monolith, serverless ## Performance & Optimization ### Relevance Scoring The system uses a sophisticated scoring algorithm: - **Time-based**: Recent interactions get higher scores (1.8x for <1 hour, 1.5x for <6 hours) - **Interaction Type**: Conversation turns (1.5x), client requests (1.3x), agent responses (1.2x) - **Content Length**: Longer content gets higher scores (1.3x for >200 chars, 1.2x for >100 chars) - **Status**: Successful interactions get 1.1x boost ### Caching & Performance - **Context Cache**: Frequently accessed contexts are cached in memory - **Database Indexes**: Optimized queries for session and topic lookups - **Lazy Loading**: Context is generated on-demand, not pre-computed ## Monitoring & Debugging ### Context Analysis Tool Use `get_conversation_context()` to debug context generation: ```python # Get detailed context breakdown context = get_conversation_context() print(context) ``` This shows: - Decision tree structure - Active conversation branches - User preference analysis - Project context extraction - Semantic analysis results ### Logging & Metrics The system logs: - Context generation success/failure - Decision tree building statistics - Context injection performance - User preference learning patterns ## Integration with Cursor ### Automatic Context Injection The system automatically enhances every interaction by: 1. **Analyzing** recent conversation history 2. **Building** decision tree structure 3. **Extracting** relevant context 4. **Injecting** context into prompts 5. **Learning** from user responses ### Custom Context Injection For specific use cases, manually inject context: ```python # Get context for specific session context = get_conversation_context(session_id="user-123") # Inject context into custom prompt enhanced_prompt = inject_conversation_context( "Help me refactor this legacy code", session_id="user-123" ) ``` ## Future Enhancements ### Planned Features - **Embedding-based Similarity**: Use vector embeddings for better semantic matching - **Machine Learning**: Learn context patterns from user feedback - **Multi-modal Context**: Support for code, images, and documents - **Context Sharing**: Share context across team members - **Context Templates**: Pre-defined context patterns for common scenarios ### Extensibility The system is designed to be easily extended: - Add new topic categories - Implement custom relevance scoring - Create specialized context extractors - Integrate with external tools and APIs ## Troubleshooting ### Common Issues **Context Not Available** ```bash # Check database connection python test_postgres.py # Verify migration completed python migrate_database.py ``` **Poor Context Quality** ```bash # Analyze context generation context = get_conversation_context() print(context) # Check interaction logging summary = get_conversation_summary() print(summary) ``` **Performance Issues** ```bash # Check database indexes # Monitor context generation time # Verify caching is working ``` ### Debug Mode Enable debug logging: ```bash export LOG_LEVEL=DEBUG python main.py ``` ## Support & Contributing ### Getting Help - Check the logs for error messages - Use `get_conversation_context()` for debugging - Verify database schema with migration script - Test individual components ### Contributing - Add new topic categories - Improve relevance scoring algorithms - Enhance context extraction patterns - Optimize performance and caching --- **The Enhanced Conversation Context Injection System transforms your Cursor agent into a context-aware, personalized coding assistant that remembers your preferences, understands your projects, and provides more relevant and helpful responses.**