README.md

Defect Solver 🪲

AI-powered bug localization for microservice architectures using hierarchical code understanding and LLM-based analysis.

⚠️ Disclaimer: Some links point to private repositories accessible only to authorized team members.

PROJECT INDEX:

1. Defect Solver API: locates buggy repos and files
2. Defect Solver Codebase Summarizer: converts codebases to NL knowledge base
3. Dnext Coder MCP: a gateway to acccess our API
4. Central Storage: stores NL knowledge bases
5. Defect Solver Agent: a simple code agent to fix bugs

What It Does

Provide a bug description in natural language, get back a ranked list of microservices and files likely containing the defect.

Example:

User: "Why is the user profile not loading?"
Defect Solver: Bug likely in user-service microservice, file UserProfileController.java

How It Works

The system uses MCP (Model Context Protocol) to integrate with AI development environments:

flowchart LR
    User -->|Bug Description| Agent
    Agent -->|MCP Tool| Server[MCP Server]
    Server -->|API Call| DS[Defect Solver API]
    DS -->|Ranked Results| Agent

Loading

Quick Start

1. Connect to MCP Server - Use hosted version at https://dnext-coder-mcp-server.pia-team.com/mcp/
2. Configure Your IDE - VSCode, JetBrains, Claude Desktop, or any MCP-compatible environment
3. Follow Setup Guide - See User Guide for configuration and usage
4. Configure AI Agent - Copy AGENTS.md to your project for optimal agent behavior

Architecture

Two-phase pipeline: Search Space Routing → Bug Localization

1. Phase 1: Identify top-N suspicious microservices from bug description
2. Phase 2: Pinpoint top-M suspicious files within selected microservices

See Algorithm Details for complete breakdown.

Development

For contributors and developers, see Developer Guide.

Research

Based on hierarchical code understanding to overcome LLM context window limitations in large-scale projects.

Paper: Repository-Level Code Understanding by LLMs via Hierarchical Summarization