qinx18/compiler-mcp-server

Compiler MCP Server

An intelligent compiler MCP server focused on vectorization analysis for C/C++ code.

Features

• Loop-carried dependency detection
• Vectorization failure analysis
• Optimization suggestions for C/C++ code
• Stateful compilation sessions
• MCP protocol compliance
• Real-time compilation feedback

Quick Start

Installation

# Install in development mode
make install-dev

# Or manually
pip install -e ".[dev]"
pre-commit install

Development

Prerequisites

• Python 3.8+
• GCC or compatible C/C++ compiler
• Git

Development Setup

# Clone and setup
git clone <repository-url>
cd compiler-mcp-server
make dev-setup

Code Quality

This project uses modern Python development practices:

• Type Checking: MyPy for static type analysis
• Linting: Ruff for fast Python linting
• Formatting: Black and Ruff for code formatting
• Security: Bandit for security vulnerability scanning
• Pre-commit Hooks: Automated quality checks
• Testing: Pytest with coverage reporting
• CI/CD: GitHub Actions for continuous integration

Available Commands

# Development workflow
make help              # Show all available commands
make quality           # Run all quality checks
make test              # Run tests with coverage
make lint              # Run linting
make format            # Format code
make type-check        # Run type checking
make security-check    # Run security analysis
make pre-commit        # Run pre-commit hooks
make clean             # Clean build artifacts
make build             # Build package

# Quick development cycle
make format lint type-check test

See for detailed development guidelines.

Usage

# Test mode
python solution_for_s1113.py --mode test

# Run as MCP server
python solution_for_s1113.py

OpenHands Resolver Setup

This repository includes an automated issue resolver using OpenHands. To enable it:

1. Set up GitHub Secrets (in repository Settings > Secrets and variables > Actions):

   • PAT_TOKEN: Personal Access Token with repo permissions
   • PAT_USERNAME: Your GitHub username
   • LLM_API_KEY: API key for Claude or other LLM service
   • LLM_MODEL: Model name (e.g., anthropic/claude-3-5-sonnet-20241022)

2. Enable Workflow Permissions:

   • Go to Settings > Actions > General > Workflow permissions
   • Select "Read and write permissions"
   • Enable "Allow GitHub Actions to create and approve pull requests"

3. Usage:

   • Add the fix-me label to any issue for automatic resolution
   • Or mention @openhands-agent in a comment

Example

The server can analyze code like this s1113 pattern:

for (int i = 0; i < LEN_1D; i++) {
    a[i] = a[LEN_1D/2 - i] + b[i];
}

And provide detailed feedback about why vectorization fails.