MCTS MCP Server

# MCTS MCP Server Usage Guide This guide explains how to effectively use the MCTS MCP Server with Claude for deep, explorative analysis. ## Setup 1. Run the setup script to prepare the environment: ```bash ./setup.sh ``` The setup script will: - Install UV (Astral UV) if not already installed - Create a virtual environment using UV - Install dependencies with UV - Create necessary state directory 2. Add the MCP server configuration to Claude Desktop: - Copy the content from `claude_desktop_config.json` - Add it to your Claude Desktop configuration file (typically at `~/.claude/claude_desktop_config.json`) - Update paths in the configuration if necessary - Restart Claude Desktop ## Using the MCTS Analysis with Claude Once the MCP server is configured, Claude can leverage MCTS for deep analysis of topics. Here are some example conversation patterns: ### Starting a New Analysis Simply provide a question, topic, or text that you want Claude to analyze deeply: ``` Analyze the ethical implications of artificial general intelligence. ``` Claude will: 1. Initialize the MCTS system 2. Generate an initial analysis 3. Run MCTS iterations to explore different perspectives 4. Find the best analysis and generate a synthesis 5. Present the results ### Continuing an Analysis To build upon a previous analysis in the same chat session: ``` Continue exploring the technological feasibility aspects. ``` Claude will: 1. Load the state from the previous analysis 2. Start a new MCTS run that builds upon the previous knowledge 3. Leverage learned approach preferences and avoid unfit areas 4. Present an updated analysis ### Asking About the Last Run To get information about the previous analysis: ``` What was the best score and key insights from your last analysis run? ``` Claude will summarize the results of the previous MCTS run, including the best score, approach preferences, and analysis tags. ### Asking About the Process To learn more about how the MCTS analysis works: ``` How does your MCTS analysis process work? ``` Claude will explain the MCTS algorithm and how it's used for analysis. ### Viewing/Changing Configuration To see or modify the MCTS configuration: ``` Show me the current MCTS configuration. ``` Or: ``` Can you update the MCTS configuration to use 3 iterations and 8 simulations per iteration? ``` ## Understanding MCTS Analysis Output The MCTS analysis output typically includes: 1. **Initial Analysis**: The starting point of the exploration 2. **Best Analysis Found**: The highest-scored analysis discovered through MCTS 3. **Analysis Tags**: Key concepts identified in the analysis 4. **Final Synthesis**: A conclusive statement that integrates the key insights ## Advanced Usage ### Adjusting Parameters You can ask Claude to modify parameters for deeper or more focused analysis: - Increase `max_iterations` for more thorough exploration - Increase `simulations_per_iteration` for more simulations per iteration - Adjust `exploration_weight` to balance exploration vs. exploitation - Set `early_stopping` to false to ensure all iterations complete ### Using Different Approaches You can guide Claude to explore specific philosophical approaches: ``` Continue the analysis using a more empirical approach. ``` Or: ``` Can you explore this topic from a more critical perspective? ``` ## Development Notes If you want to run or test the server directly during development: ```bash # Activate the virtual environment source .venv/bin/activate # Run the server directly uv run server.py # Or use the MCP CLI tools uv run -m mcp dev server.py ``` ## Troubleshooting - If Claude doesn't recognize the MCTS server, check that Claude Desktop is correctly configured and restarted - If analysis seems shallow, ask for more iterations or simulations - If Claude says it can't continue an analysis, it might mean no state was saved from a previous run - If you encounter dependency issues, try `uv pip sync requirements.txt` to ensure exact package versions