Complete Guide to Dependencies for Running Code-Graph-RAG

Code-Graph-RAG requires a minimal core of loguru, mcp, and tree-sitter, with five optional dependency groups available for testing, full language support, semantic search, Milvus integration, and AST pattern matching.

The vitali87/code-graph-rag repository structures its Python package to remain lightweight by default while offering feature-rich extras for specific use cases. All dependency declarations reside in the [project.optional-dependencies] table of pyproject.toml (lines 87–131), allowing selective installation based on your target environment.

Core Dependencies

The base installation includes only essential runtime packages needed for fundamental operation:

• loguru – Structured logging utilities
• mcp – Model Context Protocol implementation
• tree-sitter – Core Tree-Sitter library for parsing

These ship automatically when running pip install codegraph-rag. All additional functionality is gated behind optional extras to prevent bloat.

Optional Dependency Groups

The project defines five distinct extras in pyproject.toml, each targeting specific workflows or integrations.

Test Dependencies

Install the test extra to execute the repository's test suite, generate coverage reports, and run integration tests. This group includes:

• pytest, pytest-asyncio, pytest-cov, pytest-xdist
• testcontainers, libclang, filelock
• ast-grep-py, pyyaml

Tree-Sitter Full Language Support

The treesitter-full extra provides comprehensive grammar support for multi-language code graph construction. According to pyproject.toml (lines 100–114), this adds:

• tree-sitter-python
• tree-sitter-javascript, tree-sitter-typescript
• tree-sitter-rust, tree-sitter-go, tree-sitter-scala
• tree-sitter-java, tree-sitter-c, tree-sitter-cpp
• tree-sitter-lua, tree-sitter-php, tree-sitter-c-sharp, tree-sitter-dart

Without this extra, the parser relies on a limited subset of grammars.

Semantic Search Dependencies

Enable embedding-based retrieval and LLM integration with the semantic extra. As declared in pyproject.toml (lines 116–120), this installs:

• qdrant-client – Vector database client
• torch – PyTorch for tensor operations
• transformers – Hugging Face model library

These dependencies power the modules in codebase_rag/semantic/ for vector-based similarity search.

Milvus Vector Store Integration

For teams preferring Milvus over Qdrant, the milvus extra installs pymilvus[milvus-lite] as specified in pyproject.toml (lines 122–124).

AST-Grep Pattern Matching

The ast-grep extra adds structural query capabilities through ast-grep-py and pyyaml (lines 126–129), enabling pattern matching against abstract syntax trees.

Installation Commands

Install the core library without optional features:

pip install code-graph-rag

Add specific functionality using bracket notation:

# Testing tools

pip install "code-graph-rag[test]"

# Full language support

pip install "code-graph-rag[treesitter-full]"

# Semantic search stack

pip install "code-graph-rag[semantic]"

# Milvus vector store

pip install "code-graph-rag[milvus]"

# AST pattern matching

pip install "code-graph-rag[ast-grep]"

Combine multiple extras for development environments:

pip install "code-graph-rag[semantic,treesitter-full,test]"

Conditional Import Architecture

The library implements lazy loading to prevent import errors when extras are missing. In codebase_rag/cli.py, the CLI entry point checks for available dependencies before importing modules from codebase_rag/semantic/ or codebase_rag/parsers/.

This design ensures that:

• Core functionality remains available without heavy ML frameworks
• Import errors only occur when accessing specific features that require missing dependencies
• Container images stay small unless extended language support is explicitly requested

Summary

• Core dependencies (loguru, mcp, tree-sitter) always install automatically
• test extra provides pytest suite and coverage tools for CI/CD pipelines
• treesitter-full adds 13 Tree-Sitter grammars for multi-language code graph generation
• semantic enables Qdrant, PyTorch, and Transformers for vector-based RAG
• milvus offers an alternative vector database to Qdrant
• ast-grep supports structural code pattern matching
• All extras are declared in pyproject.toml (lines 87–131) and installable via pip bracket syntax

Frequently Asked Questions

What are the minimum dependencies required to run code-graph-rag?

The minimal installation requires only loguru, mcp, and tree-sitter. These support basic code parsing and logging functionality without requiring PyTorch, Hugging Face, or additional grammar packages. This lightweight core suits environments where you need only specific language support or basic graph construction.

How do I install support for multiple programming languages?

Add the treesitter-full extra to gain support for 13 additional languages including Rust, Go, Java, C, C++, and TypeScript. Without this extra, the parser operates with limited grammar support. Install via pip install "code-graph-rag[treesitter-full]" as defined in pyproject.toml lines 100–114.

Which dependencies are needed for semantic search and RAG?

Semantic retrieval requires the semantic extra, which installs qdrant-client, torch, and transformers. These power the embedding generation and vector similarity search features located in codebase_rag/semantic/. For alternative vector storage, add the milvus extra for pymilvus support.

Can I use Milvus instead of Qdrant for vector storage?

Yes. While the semantic extra defaults to Qdrant via qdrant-client, you can install pymilvus[milvus-lite] using the milvus extra. The codebase supports conditional imports for different vector backends, though you must ensure the appropriate client library is present in your environment.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}