Codex MCP Server

A Model Context Protocol (MCP) server for local repository indexing and code understanding.

Features

• 🔍 Semantic Code Search: Vector-based search across your entire codebase

• 🏗️ AST Analysis: Deep code structure understanding using Tree-sitter

• 📊 Real-time Indexing: File system monitoring for incremental updates

• 🔗 Dependency Tracking: Import/export relationship mapping

• 🎯 Symbol Resolution: Find definitions, references, and usages

• 🚀 VS Code Integration: Seamless integration with coding agents

Related MCP server: CodeAnalysis MCP Server

Architecture

┌─── MCP Server ───────────────────────────────────────┐
│  ├── Tools (search_code, find_symbol, etc.)         │
│  ├── Database Layer (SQLite + Vector Storage)       │
│  ├── Indexing Engine (AST + Embeddings)             │
│  ├── File System Monitor (Real-time updates)        │
│  └── Language Parsers (TypeScript, Python, etc.)   │
└─────────────────────────────────────────────────────┘

📋 Design Documents: Before making architectural changes, please review the design documentation in the design/ folder:

• System Architecture - Complete system design and component interactions

• API Contracts - MCP tool specifications and database schema

Installation

npm install
npm run build

Development

npm run dev

Usage

Configure in your MCP client (e.g., VS Code):

{
  "mcpServers": {
    "codex": {
      "command": "node",
      "args": ["path/to/codex-mcp-server/dist/index.js"],
      "env": {
        "REPO_PATH": "/path/to/your/repository"
      }
    }
  }
}

Available Tools

• search_code - Semantic search across the codebase

• find_symbol - Find function/class definitions

• get_references - Find all symbol usages

• analyze_dependencies - Show import/dependency graphs

• get_context - Get relevant context for code understanding

• index_repository - Manually trigger repository indexing

Configuration

Set environment variables:

• REPO_PATH - Path to the repository to index

• OPENAI_API_KEY - For embedding generation (optional, can use local embeddings)

• DB_PATH - Database storage path (default: ./data)

🤖 GitHub Copilot Integration

This project is designed to work seamlessly with GitHub Copilot coding agents. For detailed implementation instructions using GitHub Copilot, see GITHUB_COPILOT_GUIDE.md.

Quick Start with GitHub Copilot

1. Use the hashtag trigger: Add #github-pull-request_copilot-coding-agent to your requests

2. Provide clear context: "Implement the DatabaseManager class with SQLite integration for code indexing"

3. Reference the architecture: Point to the existing interfaces and types for consistent implementation

4. Test incrementally: Build and test each component as you implement it

Example Copilot Prompts

Implement the LanguageParser class that can parse JavaScript and TypeScript files 
to extract function definitions, class declarations, and import statements.
#github-pull-request_copilot-coding-agent

Create comprehensive unit tests for the SearchEngine class, focusing on semantic 
search functionality and vector similarity calculations.
#github-pull-request_copilot-coding-agent

For complete implementation guidance, phase-by-phase instructions, and best practices for using GitHub Copilot with this codebase, see the GitHub Copilot Guide.

License

MIT