CodeGraph RAG Tutorial: Query and Edit Your Codebase with MCP + AI

CodeGraph RAG does two jobs: it answers questions about your codebase and it lets AI assistants edit code with full context. In this tutorial, you will index a repository and query it — both from the CLI and through an MCP client like Claude Code.

Step 1: Start the stack

You need Docker for Memgraph and Qdrant, plus ripgrep and cmake as prerequisites. Then pull the official image:

docker pull ghcr.io/vitali87/code-graph-rag:latest

Run the CLI against any repository. The indexer parses source files with tree-sitter and builds the graph incrementally:

code-graph-rag index --path ./my-monorepo

Step 2: Query in natural language

Once indexed, ask questions directly. The graph layer resolves relationships that vector search alone would miss:

code-graph-rag query "Which modules import the auth package?"

Results include exact file paths and line numbers — useful for onboarding, refactoring impact analysis, and documentation generation.

Step 3: Connect it to your AI assistant

CodeGraph RAG runs as an MCP server, so Claude Code, Cursor, and other MCP-compatible assistants can query and edit your codebase directly. Configure it as an MCP endpoint and your assistant suddenly has a precise map of the whole monorepo:

Step 4: Export and share

The CLI can export the graph for visualization, letting teams browse the architecture as an interactive diagram instead of digging through folders.

Final verdict

CodeGraph RAG is one of the strongest local-first code intelligence tools available right now. The hybrid graph+vector design is genuinely different from naive chunked RAG, the MCP integration is polished, and the permissive MIT license makes it a no-brainer for teams that care about code privacy. With 4.6K+ stars and active development, it is a solid foundation for your AI-assisted development workflow.