Code-Graph RAG turns a codebase — including large multi-language monorepos — into a knowledge graph that you and your AI coding assistant can query in natural language. It parses source code with Tree-sitter, stores the structure in a Memgraph graph database, and exposes the graph through an MCP server so tools like Claude Code can understand, search and even edit code with full structural context. The MIT-licensed project has more than 5,200 stars on GitHub and is under very active development in 2026.
Plain text RAG struggles with code: chunking files into embeddings loses the relationships — who calls what, what inherits from what, what depends on what — that developers actually reason about. Code-Graph RAG instead builds a graph of functions, classes, modules and their relationships across the whole repository, across languages, and answers questions by traversing that structure. Queries like “what breaks if I change this function” or “show me everything that uses this API” become graph lookups rather than guesswork, and because it ships as an MCP server, AI assistants gain the same structural awareness while they work, which improves their edits in large codebases. The project supports monorepo layouts, incremental updates as code changes, and a wide range of languages through Tree-sitter grammars. It is a good example of where code intelligence is heading: from flat text search toward queryable, structured understanding of an entire codebase.
Free. Code-Graph RAG is open source under the MIT license; it runs against the open-source Memgraph database and your own LLM provider.
Developers working in large or multi-language codebases who want their AI assistant — and themselves — to navigate code by structure and relationships instead of text search.