MCP Reasoner

🚀 MCP Reasoner

A systematic implementation of an MCP server for inference, specifically designed for the desktop version of Claude. It integrates Beam Search and Monte Carlo Tree Search (MCTS) capabilities.

🚀 Quick Start

The MCP Reasoner is a specialized server implementation for inference, tailored to Claude's desktop version. It combines Beam Search and Monte Carlo Tree Search (MCTS) to enhance reasoning capabilities.

✨ Features

• Dual Search Strategies:
  • Beam Search: Width is configurable.
  • Monte Carlo Tree Search (MCTS): Ideal for complex decision spaces.
• Thought Scoring and Evaluation: Assesses the quality of reasoning.
• Tree-based Inference Path: Organizes the reasoning process.
• Statistical Analysis of Inference Process: Tracks performance.
• Compliance with MCP Protocol: Ensures compatibility.

📦 Installation

git clone https://github.com/Jacck/mcp-reasoner.git
cd mcp-reasoner
npm install
npm run build

📚 Documentation

Configuration

Add the following to the Claude desktop version configuration:

{
  "mcpServers": {
    "mcp-reasoner": {
      "command": "node",
      "args": ["path/to/mcp-reasoner/dist/index.js"]
    }
  }
}

Search Strategies

Beam Search

• Maintains a fixed-width set of most likely paths.
• Suitable for step-by-step reasoning.
• Best for: Math problems, logic puzzles.

Monte Carlo Tree Search

• Explores the decision space through simulations.
• Balances exploration and exploitation.
• Best for: Complex problems with uncertain outcomes.

⚠️ Important Note

Monte Carlo Tree Search enables Claude to perform excellently in the Arc AGI benchmark test (scoring 6/10), while Beam Search scores (3/10) on the same puzzles. For extremely complex tasks, it is recommended to guide Claude to use the MCTS strategy over Beam Search.

🔧 Technical Details

1. Search Strategy Selection:
   • Beam Search: Evaluates and ranks multiple solution paths.
   • Monte Carlo Tree Search: Uses UCT for node selection and random rollouts for exploration.
2. Thought Scoring is based on:
   • Level of detail.
   • Mathematical expressions.
   • Logical connectives.
   • Strength of parent-child relationships.
3. Process Management:
   • Tree-based state tracking.
   • Statistical analysis of the inference process.
   • Progress monitoring.

💻 Usage Examples

Use Cases

• Math problems
• Logic puzzles
• Step-by-step analysis
• Complex problem decomposition
• Decision tree exploration
• Strategy optimization

Future Implementations

• Implement new algorithms:
  • Iterative Deepening Depth-First Search (IDDFS)
  • Alpha-Beta Pruning

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.