Code Search MCP

🚀 Code Search MCP Server

A high - performance, batch - oriented MCP (Model Context Protocol) code understanding toolkit for AI agents, specially optimized for Java.

This toolkit is designed to address the challenge that AI agents may get "lost" in large codebases or lack precise context. It emphasizes deep parsing, parallel batch processing, and panoramic context to efficiently explore large codebases. It significantly reduces token usage while enhancing logical understanding.

🚀 Quick Start

The Code Search MCP Server offers a high - performance solution for AI agents to understand code. It is optimized for Java and can help AI agents better handle large codebases.

✨ Features

Use Cases

• Panoramic Context: Read multiple files in one call with dependency context auto - expanded.
• Structural Mapping: Build project outlines fast with deep Java annotation awareness.
• Precise Surgery: Precisely locate classes/methods/definitions and return in batch.

Core Tools

Design Notes

• stdio transport: JSON - RPC 2.0 via standard I/O.
• Deterministic protocol: absolute paths only, no path wildcards.
• Java outline enhancement: annotation backtracking merged into signatures.

Java Spring Deep Fit

More than just text search, it understands business logic:

• Layer - Aware Ordering: Intelligent sorting (Controller → Service → Impl → MQ → Repository) so AI understands the business flow immediately.
• Deep Annotation - Awareness: Instead of just seeing signatures, AI sees merged annotations (like @Transactional, @PreAuthorize) to understand business semantics.
• Smart DI Parsing: Automatically identifies injected fields (dependencies), eliminating the need for AI to guess where userService comes from.
• Project Structure: Full support for src/main/java and multi - module Maven/Gradle projects.

📦 Installation

Requirements

• Node.js v18.0.0 or later

Install & Build

npm install
npm run build

💻 Usage Examples

Integration

"mcpServers": {
  "code-search": {
    "command": "node",
    "args": ["{file}/code-search/index.js"]
  }
}

📚 Documentation

Tips

• Prefer view_files_full_context for full context in one call.
• Absolute paths are required for stability and reproducibility.
• Complex queries should be split into precise calls.

📄 License

This project is open - sourced under the MIT License. You may use, modify, and distribute it under the license terms.