Show HN: Sourcerer – MCP for semantic code search that reduces token waste
11 days ago
- #code-indexing
- #AI-agents
- #semantic-code-search
- MCP server for semantic code search & navigation helps AI agents work efficiently by reducing token usage.
- Requires OpenAI API Key for generating embeddings and Git for repository management.
- Installation options include `go install`, `brew install`, and adding to `claude mcp`.
- Sourcerer builds a semantic search index using Tree-sitter to parse source files into ASTs.
- Extracts meaningful chunks (functions, classes, methods) with stable IDs and contextual summaries.
- Automatically watches for file changes, respects `.gitignore`, and re-indexes changed files.
- Uses chromem-go for persistent vector storage and OpenAI's API for semantic similarity.
- Provides functions like `semantic_search`, `get_chunk_code`, `find_similar_chunks`, and `index_workspace`.
- Currently supports Go, JavaScript, Markdown, Python, TypeScript; planned support for C, C++, Java, Ruby, Rust.
- Open for contributions; see CONTRIBUTING.md.