Logic-LM MCP Server

A Model Context Protocol (MCP) server that provides symbolic reasoning capabilities using Logic-LM framework and Answer Set Programming (ASP).

Attribution

This implementation is inspired by and builds upon the Logic-LLM framework:

Original Research:

• Paper: Logic-LLM: Empowering Large Language Models with Symbolic Solvers for Faithful Logical Reasoning
• Repository: teacherpeterpan/Logic-LLM
• Authors: Liangming Pan, Alon Albalak, Xinyi Wang, William Yang Wang

This MCP server adapts the Logic-LLM approach for integration with Claude Code and other MCP clients, providing LLM-collaborative symbolic reasoning through Answer Set Programming.

🚀 Quick Start

Prerequisites

• Python 3.8 or higher

Installation

Choose your preferred installation method:

Option 1: Install from PyPI (Recommended) ✅ LIVE ON PYPI

# Install with pip
pip install logic-lm-mcp-server

# Or install with uv (10-100x faster)
uv pip install logic-lm-mcp-server

📦 Package URL: https://pypi.org/project/logic-lm-mcp-server/

Option 2: Install with Clingo solver (for full functionality)

# Install with optional solver
pip install logic-lm-mcp-server[solver]

# Or with uv
uv pip install logic-lm-mcp-server[solver]

Option 3: Development Installation

git clone https://github.com/stevenwangbe/logic-lm-mcp-server.git
cd logic-lm-mcp-server
pip install -e .

Test Installation

logic-lm-mcp --help

Integration with Claude Code

After installing the package, add it to your Claude Code configuration:

Method 1: Using the console command (after PyPI installation)

claude mcp add logic-lm-mcp logic-lm-mcp

Method 2: Manual configuration

Edit ~/.config/claude/claude_desktop_config.json (create if it doesn't exist):

{
  "mcpServers": {
    "logic-lm": {
      "command": "logic-lm-mcp"
    }
  }
}

Restart Claude Code to load the new MCP server.

1. Test the integration:

Try these commands in Claude Code:

Check Logic-LM server health
Translate this logic problem to ASP: "All birds can fly. Penguins are birds. Can penguins fly?"

Alternative Integration (Other MCP Clients)

For other MCP-compatible tools, start the server manually:

python start_server.py

The server will run on stdio and provide these tools:

• get_asp_guidelines - Get ASP translation guidelines
• translate_to_asp_instructions - Get problem-specific ASP guidance
• verify_asp_program - Execute ASP programs with Clingo
• check_solver_health - Verify system health

Overview

Logic-LM MCP Server converts natural language logical problems into Answer Set Programming (ASP) format, solves them using the Clingo solver, and returns human-readable results. It provides a three-stage reasoning pipeline: Problem Formulation → Symbolic Reasoning → Result Interpretation.

Features

• Natural Language Input: Convert English logical problems to formal representations
• ASP-Based Reasoning: Uses Answer Set Programming for robust logical inference
• Clingo Integration: Leverages the Clingo ASP solver for symbolic reasoning
• Self-Refinement: Iterative improvement of solutions through multiple reasoning passes
• Template Library: Reusable ASP patterns for common logical structures
• Fallback Handling: Graceful degradation when solver components unavailable
• FastMCP Integration: Modern MCP server implementation with type safety

Tools Provided

1. get_asp_guidelines

Get comprehensive ASP translation guidelines (cached for efficiency).

Parameters: None

Returns: Complete ASP Logic Translation Guidelines document with comprehensive instructions for translating natural language into Answer Set Programming format.

2. translate_to_asp_instructions

Get lightweight instructions for translating a specific natural language problem to ASP.

Parameters:

• problem (string, required): Natural language logical problem to translate

Example:

{
  "problem": "All cats are mammals. Fluffy is a cat. Is Fluffy a mammal?"
}

Response:

{
  "success": true,
  "solution": "TRANSLATE TO ASP: All cats are mammals...

INSTRUCTIONS:
1. Call get_asp_guidelines() for complete patterns
2. Analyze logical structure...",
  "confidence": 1.0,
  "method": "lightweight_translation_instructions",
  "metadata": {
    "problem_length": 58,
    "guidelines_cached": false,
    "next_steps": ["Call get_asp_guidelines() if needed", "Generate ASP code", "Call verify_asp_program()"]
  }
}

3. verify_asp_program

Directly verify and solve an ASP program using the Clingo solver.

Parameters:

• program (string, required): ASP program code to verify and solve
• max_models (integer, 1-100, default: 10): Maximum number of models to find

Example:

{
  "program": "% Facts
cat(fluffy).

% Rule: All cats are mammals
mammal(X) :- cat(X).

% Query
#show mammal/1.",
  "max_models": 10
}

4. check_solver_health

Check Logic-LM server and Clingo solver health status.

Returns:

• Server status and component initialization status
• Clingo availability and version information
• System capabilities and configuration details
• Basic functionality test results

Architecture

Core Components

1. LogicFramework: Main reasoning orchestrator
2. ClingoSolver: ASP solver interface and management
3. ASPTemplateLibrary: Reusable logical pattern templates
4. FastMCP Integration: Modern MCP server implementation

Processing Pipeline

Natural Language Input
         ↓
LLM Translation Instructions (Problem-specific guidance)
         ↓  
ASP Program Generation (LLM-driven with guidelines)
         ↓
Clingo Solver Execution
         ↓
Model Interpretation (Symbolic results)
         ↓
Human-Readable Output

Dependencies

• Python 3.8+: Core runtime environment
• FastMCP 2.0+: Modern MCP server framework
• Pydantic 2.0+: Input validation and type safety
• Clingo 5.8.0+: ASP solver (automatically detects if missing)

Installation

Option 1: Using pip

pip install -r requirements.txt

Option 2: Manual installation

pip install fastmcp>=2.0.0 pydantic>=2.0.0 clingo>=5.8.0

Option 3: Development setup

git clone <repository-url>
cd logic-lm-mcp-server
pip install -e .

Configuration

The server automatically handles:

• Clingo solver installation detection
• Template library loading
• Environment-specific optimizations
• Error recovery and fallback modes

Environment Variables

• No environment variables required
• Server runs with sensible defaults

Usage Examples

Basic Logical Reasoning

Input: "If it's raining, then the ground is wet. It's raining. Is the ground wet?"
Output: "Yes, the ground is wet. This conclusion follows from modus ponens..."

Syllogistic Reasoning

Input: "All birds can fly. Penguins are birds. Can penguins fly?"
Output: "Based on the given premises, yes. However, this conflicts with real-world knowledge..."

Set-Based Logic

Input: "All members of set A are in set B. X is in set A. Is X in set B?"
Output: "Yes, X is in set B. This follows from set inclusion transitivity..."

Testing

Basic Functionality Test

logic-lm-mcp --help

Test MCP Integration

# Test with Claude Code
claude mcp get logic-lm

Error Handling

• Clingo Unavailable: Provides informative error messages with installation guidance
• Invalid ASP Programs: Syntax checking with detailed error messages
• Solver Timeouts: Graceful handling of complex problems
• Resource Constraints: Memory and time limit management

Performance

• Simple Problems: 50-200ms response time
• Complex Reasoning: 200-1000ms with self-refinement
• Memory Usage: ~25MB base + ~1MB per concurrent request
• Concurrent Support: Multiple simultaneous reasoning requests

Troubleshooting

Common Issues

1. "No module named 'pydantic'" or similar

   • Install dependencies: pip install -r requirements.txt

2. "Clingo not available"

   • Install Clingo: pip install clingo
   • Server will run with limited functionality if Clingo is missing

3. Server fails to start

   • Check Python version: python --version (requires 3.8+)
   • Test installation: logic-lm-mcp --help

4. MCP connection issues

   • Verify MCP server configuration: claude mcp get logic-lm
   • Check installation: logic-lm-mcp --help

Getting Help

1. Test installation: logic-lm-mcp --help
2. Check the health endpoint: use check_solver_health tool
3. Enable debug traces: set include_trace=true in requests

FAQ - Common Setup Errors

"Missing required dependencies" on startup

Error:

❌ Missing required dependencies:
  - fastmcp>=2.0.0
  - pydantic>=2.0.0

Cause: Dependencies not properly installed or virtual environment not activated.

Solution:

# Option 1: Use virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

# Option 2: Install globally
pip install -r requirements.txt

# Option 3: Use venv python directly
venv/bin/python start_server.py

"ModuleNotFoundError: No module named 'fastmcp'"

Error:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'fastmcp'

Cause: Virtual environment not properly activated or dependencies not installed.

Solution:

# Clean installation
rm -rf venv/
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

"ModuleNotFoundError: No module named 'pydantic'"

Error:

ModuleNotFoundError: No module named 'pyda

---