Logic-LM MCP Server

# Logic-LM MCP Setup Guide This guide helps you set up the Logic-LM MCP server for optimal user experience, including automatic formatting of logical reasoning outputs. ## Quick Setup ### 1. Install and Start Server ```bash # Install dependencies pip install -r requirements.txt # Start the MCP server python start_server.py ``` ### 2. Add to Claude Code MCP Configuration Add this server to your Claude Code configuration: ```json { "mcpServers": { "logic-lm": { "command": "python", "args": ["path/to/logic-lm-mcp-server/start_server.py"], "cwd": "path/to/logic-lm-mcp-server" } } } ``` ### 3. Test the Setup ```bash # Test basic functionality mcp__logic-lm__check_solver_health # Test reasoning mcp__logic-lm__translate_to_asp_instructions "All cats are mammals. Fluffy is a cat. Is Fluffy a mammal?" ``` ## Enhanced User Experience (Recommended) For the best experience with logical reasoning outputs, add these instructions to your global `~/.claude/CLAUDE.md` file: ### Option A: Automatic Logic-LM Formatting Add this section to your `~/.claude/CLAUDE.md`: ```markdown # Logic-LM Automatic Formatting When using Logic-LM MCP tools, automatically format responses for readability: ## Logic Reasoning Response Format When Logic-LM tools return symbolic reasoning results, format the response as: **Problem**: [Restate the original logical problem clearly] **Reasoning**: [Explain the logical steps in natural language] **Conclusion**: [State the final answer clearly] **ASP Code** (if generated): ```asp [Show the Answer Set Programming code] ``` **Solver Results**: [Interpret the symbolic solver output] This applies to: - `translate_to_asp_instructions` responses - `verify_asp_program` results - Any logical reasoning workflow Example format: > **Problem**: All cats are mammals. Fluffy is a cat. Is Fluffy a mammal? > > **Reasoning**: This is a classic syllogistic reasoning pattern. We have a universal rule (all cats are mammals) and a specific instance (Fluffy is a cat). By applying the universal rule to the specific case, we can deduce the conclusion. > > **Conclusion**: Yes, Fluffy is a mammal. > > **ASP Code**: > ```asp > cat(fluffy). > mammal(X) :- cat(X). > #show mammal/1. > ``` > > **Solver Results**: The Clingo solver found one model with `mammal(fluffy)`, confirming our logical deduction. ``` ### Option B: Manual Formatting If you prefer to control formatting manually, you can format Logic-LM outputs by: 1. Using `get_asp_guidelines()` once to understand ASP translation patterns 2. Using `translate_to_asp_instructions()` for problem-specific guidance 3. Manually creating ASP code based on the instructions 4. Using `verify_asp_program()` to validate and solve the ASP code 5. Interpreting results in natural language ## Advanced Configuration ### Environment Variables No special environment variables are required. The server auto-detects: - Python version compatibility - Clingo solver availability - MCP framework version ### Performance Optimization The server includes several optimizations: - **Guideline Caching**: ASP translation guidelines are cached after first load - **Template Library**: Pre-built ASP patterns for common logical structures - **Lightweight Instructions**: Problem-specific guidance uses minimal context ### Context Management To minimize token consumption: - `get_asp_guidelines()` returns full guidelines (~9,174 chars) - call once per session - `translate_to_asp_instructions()` returns lightweight guidance (~500 chars) - call for each problem - Guidelines are cached automatically to prevent repeated large context injections ## Workflow Examples ### Basic Logical Reasoning ```bash # Get problem-specific ASP translation guidance mcp__logic-lm__translate_to_asp_instructions "If it's raining, then the ground is wet. It's raining. Is the ground wet?" # Generate ASP code based on instructions (done by LLM/user) # cat(fluffy). # mammal(X) :- cat(X). # #show mammal/1. # Verify and solve the ASP program mcp__logic-lm__verify_asp_program "% Facts\nraining.\n\n% Rule\nwet_ground :- raining.\n\n% Query\n#show wet_ground/0." ``` ### Complex Multi-Step Reasoning ```bash # For complex problems, get full guidelines first mcp__logic-lm__get_asp_guidelines # Then get problem-specific instructions mcp__logic-lm__translate_to_asp_instructions "All birds can fly. Penguins are birds. Penguins cannot fly. Resolve this contradiction." # Work through the logical analysis with full guideline context # Generate appropriate ASP code with constraints # Verify the program handles the contradiction properly ``` ## Troubleshooting ### Common Issues 1. **Server won't start**: Check Python version (3.8+ required) and dependencies 2. **Clingo not found**: Install with `pip install clingo` or `conda install -c conda-forge clingo` 3. **MCP connection issues**: Verify the server path in Claude Code configuration 4. **Guidelines not loading**: Check that `docs/asp_logic_guidelines.md` exists in the server directory ### Health Check Use the health check tool to verify everything is working: ```bash mcp__logic-lm__check_solver_health ``` This will report: - Server status and initialization - Clingo solver availability and version - Component health and capabilities - Recommendations for any issues found ### Getting Help 1. Check the main [README.md](README.md) for detailed documentation 2. Run the health check tool for system diagnostics 3. Test with simple logical problems first 4. Verify ASP code syntax if solver reports errors ## Integration Tips ### With Other MCP Servers Logic-LM works well alongside other MCP servers for comprehensive problem-solving workflows. ### With Symbolic Solver Integration If you have symbolic solver integration in your CLAUDE.md, Logic-LM complements it by providing: - ASP-specific reasoning capabilities - Template-based logical pattern matching - Clingo solver access for Answer Set Programming ### Custom Templates You can extend the ASP template library by editing `src/asp_templates.py` and adding new patterns for domain-specific logical structures. --- ## Next Steps 1. ✅ Complete basic setup and test functionality 2. ✅ Add CLAUDE.md formatting instructions for better UX 3. ✅ Try the example reasoning problems 4. ✅ Explore complex logical scenarios 5. ✅ Integrate with your existing reasoning workflows For questions or issues, refer to the troubleshooting section or check the health status of your Logic-LM installation.