MCP Reasoner

The MCP Reasoner provides a configurable server for Claude Desktop that combines two powerful search strategies—Beam Search and Monte Carlo Tree Search (MCTS)—to perform structured, step-by-step reasoning and complex decision making. It includes thought scoring, tree-based reasoning paths, and statistical analysis to help you understand and tune the reasoning process while staying MCP protocol-compliant.

How to use

Connect your Claude Desktop client to the MCP Reasoner server to leverage dual search strategies for step-by-step analysis and deep decision-making. You can switch between Beam Search for straightforward, orderly reasoning and Monte Carlo Tree Search for exploring uncertain or large decision spaces. The server tracks reasoning paths, scores thoughts, and provides statistical insights to help you monitor progress and adjust strategies.

How to install

Prerequisites: ensure you have Node.js and npm installed on your system.

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

Configuration and usage notes

To run the MCP Reasoner with Claude Desktop, configure the MCP server in your Claude Desktop setup to point to the local Node.js runtime and the built entry point. Use the following configuration snippet to connect via a stdio-based approach.

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

Search strategies and when to use them

Beam Search keeps a fixed-width set of the most promising reasoning paths. It is ideal for structured, step-by-step deduction, useful for mathematical problems and logical puzzles.

Monte Carlo Tree Search explores the decision space through simulations, balancing exploration and exploitation. It is well-suited for complex problems with uncertainty and large search spaces.

Tip: For super complex tasks, prefer MCTS to guide Claude through ambiguous or highly variable reasoning paths.

Algorithm details

1. Search Strategy Selection: Beam Search evaluates and ranks multiple solution paths while MCTS uses node selection and random rollouts to explore possibilities.

2. Thought Scoring: Each thought is scored based on detail level, mathematical expressions, logical connectors, and the strength of parent-child relationships.

3. Process Management: The system tracks state in a tree structure and provides statistical analysis and progress monitoring.

Use cases

Mathematical problems, logical puzzles, step-by-step analyses, complex problem decomposition, and decision tree exploration.

Strategy optimization and structured reasoning for challenging tasks.

Future implementations

Potential enhancements include Iterative Deepening Depth-First Search (IDDFS) and Alpha-Beta Pruning to improve performance and pruning efficiency.

Notes on licensing

This project is licensed under the MIT License.

Available tools

BeamSearch

Maintains a fixed-width set of the most promising reasoning paths to enable orderly, step-by-step deduction.

MCTS

Monte Carlo Tree Search explores the decision space through simulations, balancing exploration and exploitation for complex, uncertain problems.

ThoughtScoring

Assesses the quality of each thought based on detail level, mathematical content, logical connectors, and parent-child relationship strength.

TreeBasedReasoning

Keeps reasoning states in a tree structure to enable traceable, hierarchical analysis.

StatisticalAnalysis

Generates analytics on the reasoning process to help you monitor progress and tune strategies.

MCPCompliance

Ensures interactions adhere to the MCP protocol for interoperability.