Mcp-Reasoning

This MCP server provides advanced reasoning capabilities for Cursor AI, including Monte Carlo Tree Search (MCTS), beam search, R1 Transformer reasoning, and hybrid approaches that combine transformer analysis with search strategies. It is designed to be added to Cursor AI as a global MCP server so Claude can invoke specialized reasoning tools directly within conversations. Once the server is running, Cursor AI loads the available tools and exposes commands such as /reason-mcts, /reason-beam, /reason-r1, and /reason-hybrid for interactive reasoning tasks. The server is built to auto-complete reasoning steps within a single tool call, giving you an end-to-end reasoning chain and a final result without requiring multiple prompts.

To use the tools, first ensure the server is started and accessible to Cursor AI. In Cursor’s MCP settings, add a new global MCP server and point it to the Node entry file (dist/index.js) of the reasoning server. After enabling the server and refreshing tools, you can invoke the reasoning tools directly in chat using the dedicated commands:

• /reason-mcts Your query here to trigger MCTS-based reasoning
• /reason-beam Your query here to trigger beam search reasoning
• /reason-r1 Your query here to trigger single-step Transformer reasoning
• /reason-hybrid Your query here to trigger hybrid Transformer+MCTS reasoning Each command returns the full reasoning trace and the final answer, streamlining complex problem solving within Cursor AI.

Prerequisites:

• Node.js (v14+ recommended) installed on your system
• Access to a command line or terminal
• Permissions to run and host the MCP server (as needed by your environment)

1. Install dependencies and build

   • Navigate to the server project directory
   • Install dependencies npm install
   • Build the project npm run build

2. Run the server locally (example)

   • Start the server so Cursor AI can connect to it via the dist/index.js entry point node dist/index.js

3. Configure Cursor AI to use the MCP server

   • In Cursor AI, go to Settings > Features > MCP
   • Add a new global MCP server with the following details: Command: node Args: ["C:\path\to\mcp-reasoning-server\dist\index.js"]
   • Save and enable the server, then refresh the tools list

4. Validate operation

   • Open Cursor AI and issue one of the reasoning commands (e.g., /reason-mcts) to ensure the server responds with reasoning traces and results.

Tips and notes:

• The server exposes multiple reasoning tools: MCTS (/reason-mcts), beam search (/reason-beam), R1 transformer (/reason-r1), and hybrid (/reason-hybrid). Each tool performs iterative reasoning internally and returns the full chain along with the final answer.
• Changes to server code require a rebuild (npm run build) before restarting the server to pick up updates.
• Ensure the dist/index.js path is correct in your Cursor AI configuration. Use full Windows-style paths if running on Windows Desktop, with doubled backslashes as shown in the examples.
• Keep the server running in a dedicated process or window; the tools are only available while the server process is active.
• If you encounter connection issues, verify Node.js is installed, the server is built, and the path in Cursor AI points to the correct dist/index.js file.
• You can customize environment variables in the mcp_config if you need to tweak production settings or memory usage.