Ragex - Hybrid Retrieval-Augmented Generation for multi-language codebases.

An MCP (Model Context Protocol) server that analyzes codebases using compiler output and language-native tools to build comprehensive knowledge graphs for natural language querying, static analysis, and safe code editing.

Architecture

• MCP Server: JSON-RPC 2.0 server over stdio and socket (50+ tools)
• Language Analyzers: Pluggable parsers for Elixir, Erlang, Python, Ruby, JS/TS
• Knowledge Graph: ETS-based storage for code entities and relationships
• Hybrid Retrieval: Reciprocal Rank Fusion of symbolic + semantic search
• Bumblebee Embeddings: Local ML model for semantic code search
• Code Analysis: Quality, security, business logic, duplication, dead code
• Code Editing: Atomic file edits, semantic refactoring, multi-file transactions
• AI Agent: ReAct-loop agent using Ragex MCP tools for RAG (chat, audit)

Usage as MCP Server

Build and run the server:

mix compile
mix run --no-halt

The server listens for MCP messages on stdin and responds on stdout. A socket server also starts on a configurable port for persistent connections.

Interactive CLI Tasks

# AI-powered codebase Q&A via agent with Ragex MCP tools
mix ragex.chat

# AI-powered static analysis + audit report
mix ragex.audit

# Comprehensive static analysis
mix ragex.analyze