MCP Reasoner

🚀 MCP Reasoner

An inference function implemented for Claude Desktop, supporting both Beam Search and Monte Carlo Tree Search (MCTS). Initially, this project was just a test to see if we could enhance Claude's ability to solve complex problems... and it turns out we can.

Current Version:

v2.0.0

New Features:

Two experimental inference algorithms have been added:

- `mcts-002-alpha`

    - It uses the A* search method and includes an early *alpha* implementation of the policy simulation layer.

    - It also includes an early *alpha* implementation of the adaptive exploration simulator and the result-based inference simulator.

*Note*: The implementation of these simulators is not yet complete and may change.

- `mcts-002alt-alpha`

    - It uses the bidirectional search method and includes an early *alpha* implementation of the policy simulation layer.

    - It also includes an early *alpha* implementation of the adaptive exploration simulator and the result-based inference simulator.

*Note*: The implementation of these simulators is not yet complete and may change.

What happened to mcts-001-alpha and mcts-001alt-alpha?

Simply put: they were ineffective and almost similar to the basic mcts method. After initial testing, the results showed that the basic thinking processes were similar, indicating that simply adding policy simulation might not have much effect.

So why add the policy simulation layer now?

I believe that combining both policy and search is the key, as most algorithms are implemented in this way.

Previous Version:

v1.1.0

Model control over search parameters has been added:

beamWidth - Allows Claude to adjust the number of different paths to track (1 - 10)

numSimulations - Fine-tunes the number of MCTS simulations (1 - 150)

✨ Features

• Offers two switchable search strategies:
  • Beam Search (suitable for simple problems)
  • MCTS (for complex problems) and its alpha variants (see above)
• Tracks the quality of different inference paths
• Visualizes all different ways Claude solves problems
• Analyzes the progress of the inference process
• Follows the MCP protocol (obviously)

📦 Installation

git clone https://github.com/frgmt0/mcp-reasoner.git 

OR clone the original version:

git clone https://github.com/Jacck/mcp-reasoner.git

cd mcp-reasoner
npm install
npm run build

📚 Configuration

Add to the Claude Desktop configuration:

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

💻 Testing

[More tests are coming soon]

🔧 Benchmarking

[Benchmarks will be added soon]

Key benchmark items:

• MATH500
• GPQA-Diamond
• GMSK8
• Perhaps Polyglot and/or SWE-Bench

📄 License

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