Anmol369/Pattern-Learning-MCP-Server

🎯 Pattern Learning MCP Server - Demo

A demonstration of verification time reduction through pattern learning for AI code generation

📋 Overview

This demo showcases how an MCP (Model Context Protocol) server can help AI agents like Goose learn from developer corrections, reducing verification time from 20 minutes to 2 minutes (90% reduction).

The Problem

• AI generates code with security vulnerabilities
• Developers spend significant time reviewing and fixing
• Same mistakes happen repeatedly

The Solution

• Store correction patterns when developers fix AI-generated code
• Retrieve and apply these patterns in future code generation
• Reduce verification time dramatically

🚀 Quick Start

1. Install Dependencies

cd pattern-learning-demo
pip install -r requirements.txt

2. Run the Demo

Option A: Python Script (Recommended for Windows)

python demo_script.py

Option B: Bash Script (Linux/Mac)

chmod +x demo_script.sh
./demo_script.sh

3. Test the MCP Server

python mcp_server.py

The server will start and initialize the pattern database with security patterns.

📊 Demo Flow

Before Pattern Learning (Attempt 1)

1. Developer asks: "Create Flask login endpoint"
2. Goose generates code with timing attack vulnerability (password == stored_password)
3. Developer spends 20 minutes reviewing and fixing
4. Pattern stored in database

After Pattern Learning (Attempt 2)

1. Developer asks: "Create Flask login endpoint" (same prompt)
2. Goose calls MCP server to get security patterns
3. Goose generates code with secrets.compare_digest() from the start
4. Developer verifies in 2 minutes
5. 90% time reduction achieved

🏗️ Architecture

┌─────────────────────────────────────────┐
│        GOOSE AGENT                      │
│  (calls MCP tools via JSON-RPC)        │
└───────────────┬─────────────────────────┘
                │
                ▼
┌─────────────────────────────────────────┐
│   PATTERN LEARNING MCP SERVER           │
│                                         │
│  Tools:                                 │
│  • get_security_patterns()             │
│  • validate_code_security()            │
│  • store_correction_pattern()          │
│  • get_database_stats()                │
└───────────────┬─────────────────────────┘
                │
                ▼
┌─────────────────────────────────────────┐
│     SQLite Pattern Database             │
│  (patterns.db)                          │
└─────────────────────────────────────────┘

🔧 Components

1. pattern_db.py - Database Layer

• SQLite-based pattern storage
• Bayesian confidence scoring
• Pattern retrieval and statistics

2. security_patterns.py - Pattern Definitions

• Pre-defined security patterns
• Vulnerability detection (regex-based)
• Context-aware recommendations

3. mcp_server.py - MCP Server

• Exposes tools via MCP protocol
• Handles pattern storage and retrieval
• Integrates with Goose

4. demo_script.py - Interactive Demo

• Shows before/after comparison
• Visualizes time savings
• Demonstrates the value proposition

📈 Results

🎯 Key Features

✅ Automatic Pattern Learning - Learns from every correction ✅ Context-Aware - Recommends patterns based on task type ✅ Confidence Scoring - Bayesian updates strengthen patterns over time ✅ Multiple Categories - Security, performance, style, etc. ✅ MCP Integration - Works seamlessly with Goose ✅ Zero Configuration - SQLite, no external database needed

🔒 Security Patterns Included

1. Timing Attack Prevention

   • Detects: password == stored_password
   • Fix: secrets.compare_digest(password, stored_password)

2. SQL Injection Prevention

   • Detects: String formatting in SQL queries
   • Fix: Parameterized queries

3. Hardcoded Secrets Detection

   • Detects: API_KEY = "secret123"
   • Fix: os.environ.get("API_KEY")

4. Insecure Random Prevention

   • Detects: random.randint()
   • Fix: secrets.token_hex()

🧪 Testing

# Run tests (when implemented)
pytest tests/

📝 Configuration for Goose

To use this MCP server with Goose, add to your Goose configuration:

{
  "mcp_servers": {
    "pattern-learning": {
      "command": "python",
      "args": ["C:/Users/aadis/goose-pattern-demo/pattern-learning-demo/mcp_server.py"]
    }
  }
}

🎬 Demo Script Output

The demo shows:

1. Attempt 1 - Vulnerable code generation and 20-minute fix
2. Pattern Storage - Correction saved to database
3. Attempt 2 - Secure code generated automatically
4. Results - Time savings and metrics

💡 Future Enhancements

• AST-based vulnerability detection (more accurate)
• Vector embeddings for pattern similarity search
• Multi-agent coordination
• Autonomous prompt evolution
• Community pattern sharing
• Performance pattern categories
• Style guide enforcement

📞 Support

For questions or issues:

• Email: your-email@example.com
• GitHub: your-github-username

📄 License

MIT License - Feel free to use and modify

Built for the Goose Grant Interview

Demonstrating verification time reduction through experiential learning.