MCP Codebase Index Server

AI-powered semantic search for your codebase in GitHub Copilot, Kiro, and other MCP-compatible editors

A Model Context Protocol (MCP) server that enables AI editors to search and understand your codebase using Google's Gemini embeddings and Qdrant vector storage.

Supported Editors:

• ✅ VS Code with GitHub Copilot

• ✅ VS Code with Roo Cline

• ✅ GitHub Copilot CLI

• ✅ Google Gemini CLI

• ✅ Kiro AI Editor

• ✅ Any MCP-compatible editor

📚 Quick Navigation

🚀 Getting Started

• 📖 Full Documentation - Complete documentation

• ⚙️ Setup Guide - VS Code - Installation for VS Code Copilot

• 🖥️ Setup Guide - CLI - Installation for GitHub Copilot CLI

• 🤖 Setup Guide - Gemini CLI - Installation for Google Gemini CLI

• 🎯 Setup Guide - Kiro - Installation for Kiro AI Editor

• 🦘 Setup Guide - Roo Cline - Installation for Roo Cline (VS Code)

• ⚡ Quick Reference - Command cheat sheet

• 🗺️ Navigation Guide - Find any doc quickly

💻 For Developers

• Source Code Structure - Code organization

• MCP Server Guide - Build your own MCP server

• Roadmap - Future plans

🔧 Resources

• Qdrant Setup - Get Qdrant credentials

• Testing Guide - Test search functionality

• Prompt Enhancement Guide - Use prompt enhancement effectively

• Vector Visualization Guide - Visualize your codebase

• Changelog - Version history

✨ Features

• 🔍 Semantic Search - Find code by meaning, not just keywords

• 🎯 Smart Chunking - Automatically splits code into logical functions/classes

• 🔄 Incremental Indexing - Only re-indexes changed files (90%+ time savings)

• 💾 Auto-save Checkpoints - Saves progress every 10 files, resume anytime

• 📊 Real-time Progress - Track indexing with ETA and performance metrics

• ⚡ Parallel Processing - 25x faster indexing with batch execution

• 🔄 Real-time Watch - Auto-updates index on file changes

• 🌐 Multi-language - Supports 15+ programming languages

• ☁️ Vector Storage - Uses Qdrant for persistent storage

• 🤖 Prompt Enhancement - AI-powered query improvement (optional)

• � Vector Visualization - 2D/3D UMAP visualization of your codebase

• 🏗️ Modular Architecture - Clean handler separation for maintainability

• �📦 Simple Setup - Just 4 environment variables

🚀 Quick Start

Prerequisites

1. Gemini API Key - Get free at Google AI Studio

2. Qdrant Cloud Account - Sign up free at cloud.qdrant.io

Installation

Choose your environment:

• VS Code Users: Follow steps below or see Roo Cline Setup

• Copilot CLI Users: See Copilot CLI Setup Guide

• Gemini CLI Users: See Gemini CLI Setup Guide

• Kiro Users: See Kiro Setup Guide

Step 1: Open MCP Configuration in VS Code

1. Open GitHub Copilot Chat (Ctrl+Alt+I / Cmd+Alt+I)

2. Click Settings icon → MCP Servers → MCP Configuration (JSON)

Step 2: Add this configuration to mcp.json:

Step 3: Restart VS Code

The server will automatically:

• Connect to Qdrant Cloud

• Index your codebase

• Watch for file changes

📖 Detailed instructions:

• VS Code Setup Guide

• Copilot CLI Setup Guide

📖 Usage

Search Your Codebase

Ask GitHub Copilot:

Visualize Your Codebase

Ask GitHub Copilot:

📖 Complete guide: Vector Visualization Guide

Check Indexing Status

📖 More examples: Testing Guide

📊 Vector Visualization

See your codebase in 2D/3D space - Understand semantic relationships and code organization visually.

What is Vector Visualization?

Vector visualization transforms your codebase's 768-dimensional embeddings into interactive 2D or 3D visualizations using UMAP dimensionality reduction. This allows you to:

• 🎨 Explore semantic relationships - Similar code clusters together

• 🔍 Understand architecture - See your codebase structure at a glance

• 🎯 Debug search results - Visualize why certain code was retrieved

• 📈 Track code organization - Identify modules, patterns, and outliers

Quick Start

Visualize entire codebase:

Export as HTML:

Understanding the Visualization

Colors and Clusters:

• Each color represents a semantic cluster (module/functionality)

• Points close together = similar in meaning

• Distance reflects semantic similarity

• Outliers indicate unique/specialized code

Common Cluster Patterns:

• Blue: Frontend/UI components

• Orange: API endpoints and routes

• Green: Database models and queries

• Red: Authentication and security

• Purple: Tests and validation

• Gray: Utilities and helpers

Use Cases

1. 🏗️ Architecture Understanding

   • Visualize to see module boundaries

   • Identify tightly coupled code

   • Find opportunities for refactoring

2. 🔍 Code Discovery

   • Locate related functionality visually

   • Find all code touching a feature

   • Discover cross-cutting concerns

3. 🐛 Search Debugging

   • Understand why results were retrieved

   • See semantic relationships

   • Refine queries based on visualization

4. 👥 Team Onboarding

   • Export HTML for new developers

   • Visual guide to codebase structure

   • Interactive exploration tool

5. ✅ Refactoring Validation

   • Visualize before/after refactoring

   • Verify improved code organization

   • Track architecture evolution

Performance

Tips:

• Use 2D for faster processing (40% faster than 3D)

• Limit maxVectors for large codebases

• Export HTML for offline exploration

📖 Learn More

For detailed documentation including:

• Complete tool reference

• Interpretation guide

• Technical details (UMAP, clustering)

• Troubleshooting

• Best practices

• Advanced use cases

See: Vector Visualization Guide

🎯 Prompt Enhancement (Optional)

TL;DR: Prompt enhancement is a transparent background tool that automatically improves search quality. Just ask naturally - no need to mention "enhance" in your prompts.

Quick Overview

When enabled (PROMPT_ENHANCEMENT=true), the AI automatically:

1. Enhances your search query with codebase context

2. Searches with the improved query

3. Continues with your original request (implement, fix, explain, etc.)

Good Prompts ✅

Why these work: Clear goal (find + action) → AI knows what to do

Bad Prompts ❌

Why these fail: No clear action → AI stops after search

Key Principle

Prompt enhancement is invisible infrastructure.

Just tell the AI what you want to accomplish. It will automatically use enhancement to improve search quality behind the scenes.

Think of it like autocomplete: You don't say "use autocomplete" - you just type and it helps automatically.

📖 Learn More

For detailed guide including:

• Technical details and architecture

• Configuration options

• Real-world examples (TypeScript, Python, Dart, etc.)

• Performance tips and optimization

• Troubleshooting and FAQ

• Advanced use cases

See: Prompt Enhancement Guide

🎛️ Configuration

Required Variables

Optional Variables

📖 Full configuration guide: Setup Guide

🌍 Supported Languages

Python • TypeScript • JavaScript • Dart • Go • Rust • Java • Kotlin • Swift • Ruby • PHP • C • C++ • C# • Shell • SQL • HTML • CSS

📊 Performance

📖 Performance details: Main Documentation

🐛 Troubleshooting

Server not appearing?

1. Check Copilot Chat → Settings → MCP Servers → Show Output

2. Verify all 4 env variables are set

3. Ensure REPO_PATH is absolute path

Can't connect to Qdrant?

Indexing too slow?

• Large repos take 5-10 minutes initially

• Subsequent runs only index changed files (90%+ faster)

📖 More troubleshooting: Main Documentation

📁 Project Structure

📖 Detailed structure: Project Structure | Source Code Structure

🔧 Development

Build

Run Locally

Test

📖 Development guide: Source Code Structure

🤝 Contributing

Contributions welcome! Check out:

• Improvement Plan - Roadmap

• Issues - Detailed feature docs

• Source Code - Code structure

📄 License

MIT © NgoTaiCo

📞 Support

• Issues: GitHub Issues

• Discussions: GitHub Discussions

• Email: ngotaico.flutter@gmail.com

⭐ If you find this useful, please star the repo!