MCP Self-Learning Server
A sophisticated Model Context Protocol (MCP) server that autonomously learns from interactions, optimizes performance, and continuously improves its knowledge base through pattern recognition and machine learning techniques.
🌟 Features
🧠 Autonomous Learning Engine
- Pattern Recognition: Automatically identifies and learns from interaction patterns
- Feature Extraction: Analyzes tool sequences, context, performance metrics, and semantic embeddings
- Confidence Scoring: Evaluates pattern reliability based on frequency, recency, and consistency
- Memory Consolidation: Manages short-term and long-term pattern storage
🔄 Knowledge Synchronization
- Auto-sync: Every 60 seconds between MCP servers
- Knowledge Export/Import: JSON and Markdown formats
- Pattern Merging: With deduplication
- Cross-server Learning: Through shared knowledge directory
📊 Self-Improvement Capabilities
- Performance Optimization: Identifies redundancies and bottlenecks
- Predictive Suggestions: Anticipates next actions based on learned patterns
- Error Pattern Analysis: Learns from failures to improve success rates
- Adaptive Recommendations: Generates context-aware optimizations
💾 Data Persistence
- Automatic Data Saving: Every 5 minutes with backup rotation
- Learning Data Recovery: Loads previous sessions on startup
- Export Knowledge: Multiple formats (JSON, Markdown)
- Backup System: Automatic backup creation before saves
📝 Advanced Logging
- Multi-level Logging: Debug, Info, Warn, Error with colors and emojis
- File & Console Output: Simultaneous logging to both
- Log Rotation: Prevents disk space issues
- Performance Monitoring: Tool execution times and memory usage
🚀 Quick Start
Prerequisites
Installation
- Clone/Download the Project
cd ~/saralegui-solutions-llc/shared/MCPSelfLearningServer
- Install Dependencies
- Configure Claude DesktopAdd to
~/.config/Claude/claude_desktop_config.json:{
"mcpServers": {
"self-learning": {
"command": "node",
"args": ["/home/ben/saralegui-solutions-llc/shared/MCPSelfLearningServer/mcp-self-learning-server.js"],
"env": {
"NODE_ENV": "production",
"LEARNING_MODE": "autonomous"
}
}
}
}
- Start the Server
📋 Available Commands
Development & Testing
npm run dev # Start in development mode
npm run debug # Start with debug logging
npm test # Run all tests
npm run test:unit # Run unit tests only
npm run test:integration # Run integration tests only
Monitoring & Health
npm run health # Run comprehensive health check
npm run monitor # Real-time monitoring
npm run monitor:details # Detailed monitoring with change tracking
Manual Operations
# Health check
node tools/health-check.js
# Real-time monitoring
node tools/monitor.js [--interval 5] [--details]
# Start server directly
node mcp-self-learning-server.js
analyze_pattern
Analyze and learn from interaction patterns
{
"interaction": {
"type": "tool_usage",
"input": "user input",
"output": "tool output",
"context": {},
"performance": { "duration": 100 },
"success": true
}
}
get_insights
Get current learning analytics and insights
trigger_learning
Manually trigger a learning cycle
Knowledge Management
export_knowledge
Export learned knowledge to file
{
"format": "json|markdown" // Optional, defaults to json
}
import_knowledge
Import knowledge from external source
{
"source": "file_path_or_url",
"format": "json" // Optional
}
Get optimization suggestions for specific tools
{
"tool_name": "example_tool" // Optional
}
predict_next_action
Get predictive suggestions based on current context
{
"context": {
"current_tool": "analyze_pattern",
"user_intent": "optimization"
}
}
Get detailed performance analytics
{
"tool_name": "specific_tool" // Optional, for tool-specific metrics
}
📊 Monitoring & Analytics
Health Check Results
The health check tool verifies:
- ✅ Server startup functionality
- ✅ Data persistence system
- ✅ Logging system
- ✅ Performance metrics (startup time)
Real-time Monitoring
The monitor displays:
- Learning engine status (patterns, knowledge, cycles)
- Log file metrics and activity
- System resource usage
- Change indicators showing growth over time
| Metric | Target | Excellent |
|---|
| Startup Time | <5s | <1s |
| Memory Usage | <100MB | <50MB |
| Response Time | <500ms | <100ms |
| Learning Accuracy | >70% | >90% |
🗂️ Directory Structure
MCPSelfLearningServer/
├── mcp-self-learning-server.js # Main server file
├── package.json # Dependencies and scripts
├── README.md # This file
├── data/ # Persistent learning data
│ ├── learning-engine.json # Main learning data
│ └── learning-engine.backup.json # Backup
├── logs/ # Server logs
│ └── mcp-server.log # Main log file
├── lib/ # Shared libraries
│ └── logger.js # Enhanced logging system
├── test/ # Test suites
│ ├── unit/ # Unit tests
│ └── integration/ # Integration tests
└── tools/ # Development tools
├── health-check.js # Health check tool
└── monitor.js # Real-time monitoring
🔧 Configuration
Environment Variables
| Variable | Default | Description |
|---|
NODE_ENV | production | Environment mode |
LOG_LEVEL | info | Logging level (debug/info/warn/error) |
LOG_CONSOLE | true | Enable console logging |
LOG_FILE | true | Enable file logging |
LEARNING_MODE | autonomous | Learning behavior mode |
Learning Engine Settings
- Max Memory Size: 1000 patterns in memory
- Auto-save Interval: 5 minutes
- Pattern Confidence Threshold: 0.5
- Learning Trigger: Every 100 interactions or 50 tool uses
🚨 Troubleshooting
Common Issues
- Server Won't Start
- Check Node.js version (18+ required)
- Verify all dependencies installed:
npm install - Check file permissions
- Data Not Persisting
- Verify
data/ directory permissions - Check disk space
- Review logs for errors:
tail -f logs/mcp-server.log
- High Memory Usage
- Run health check:
npm run health - Check pattern count:
npm run monitor - Consider reducing max memory size
- Slow Performance
- Enable performance logging:
npm run debug - Check system resources
- Review learning cycle frequency
Log Analysis
# View recent logs
tail -f logs/mcp-server.log
# Search for errors
grep "ERROR" logs/mcp-server.log
# Count log levels
grep -c "INFO\|WARN\|ERROR" logs/mcp-server.log
📈 Expected Learning Outcomes
- Basic pattern recognition active
- Initial knowledge base building
- Tool usage tracking enabled
Short-term (100-1000 interactions)
- Pattern confidence scores stabilizing
- First optimization recommendations
- Predictive accuracy ~50%
Long-term (1000+ interactions)
- Predictive accuracy >70%
- Response time improvements ~30%
- Comprehensive knowledge graph
- Cross-server knowledge sharing
- Self-documenting insights
🤝 Integration with Claude
Once configured, the server provides these tools in Claude:
- Pattern analysis for learning from conversations
- Performance insights for optimization
- Predictive suggestions for improved responses
- Knowledge export for documentation
- Real-time learning from every interaction
📝 License
ISC License
🆘 Support
For issues or questions:
- Run health check:
npm run health - Check logs:
tail -f logs/mcp-server.log - Review this documentation
- Check server status:
npm run monitor
Built with ❤️ for autonomous learning and continuous improvement