code-graph-rag

Code-Graph-RAG is an open-source RAG (Retrieval-Augmented Generation) system designed specifically for large monorepos with mixed programming languages. Using Tree-sitter, it parses source code into an AST, extracts functions, classes, methods, modules, and their relationships, and stores the result as a knowledge graph in Memgraph. Users interact with the codebase through a CLI or an MCP server, asking questions in plain English. The AI model converts the query into Cypher, retrieves the relevant graph nodes, and returns grounded answers. Beyond Q&A, the tool offers AST-based structural search and replace (via ast-grep), dead code detection by traversing call/reference edges, and AI-driven code optimisation against best practices. Unlike tools like GitHub Code Search or Sourcegraph, Code-Graph-RAG provides a language-agnostic graph schema that unifies all languages, making cross-language refactoring and semantic understanding significantly easier. It also integrates with MCP, so Claude Code and other agents can directly edit code with previewed diffs.