Muvon/octocode

Semantic code searcher and codebase utility

What it solves

AI assistants often struggle to understand the full context of a codebase because they treat code as flat text. They lack an understanding of how different files relate to one another, making it difficult to track dependencies, navigate imports, or remember architectural decisions across sessions.

How it works

Octocode uses tree-sitter AST parsing to extract actual code symbols (like functions and classes) rather than arbitrary text chunks. It then builds a GraphRAG knowledge graph that maps relationships such as imports, calls, and implements across the project. This structural intelligence is exposed via a Model Context Protocol (MCP) server, allowing AI agents to perform semantic searches, view code signatures, and query the dependency graph.

Who it’s for

Developers using MCP-compatible AI tools (such as Claude Desktop, Cursor, or Windsurf) who want their AI assistants to have a deep, structural understanding of their local codebase for better onboarding, refactoring, and code review.

Highlights

  • Structural Intelligence: Moves beyond standard RAG by indexing the codebase as a knowledge graph of symbols and relationships.
  • Native MCP Integration: Provides a built-in MCP server that gives AI agents direct tools for semantic and structural search.
  • Hybrid Search: Combines semantic similarity with BM25 full-text search and reranking for higher retrieval accuracy.
  • Multi-language Support: Supports a wide array of languages including Rust, Python, TypeScript, Go, C++, and Java via tree-sitter.
  • Local-first Privacy: Designed to be local-first, respecting .gitignore and offering local embedding options.