Exquisite Context for AI Agents — One-command installation of CodeWeaver's semantic code search MCP server.
/plugin install codeweaver@knitliThat's it! CodeWeaver is now available in your Claude Code session.
- Hybrid Search: Semantic + AST + keyword search for precise code discovery
- 166+ Languages: Language-aware chunking and understanding
- 27 Languages with AST: Deep semantic analysis for major languages
- Automatic Indexing: Background indexing with file watching
- Zero Configuration: Works out of the box with sensible defaults
Once installed, CodeWeaver runs as an MCP server that Claude can query to find relevant code in your project. It combines:
- Semantic Search: Understanding what code means, not just what it says
- AST Analysis: Structural understanding of code organization
- Keyword Matching: Traditional search when appropriate
- Intelligent Ranking: RRF (Reciprocal Rank Fusion) for best results
CodeWeaver automatically indexes your project on first use. Claude can then search your codebase with natural language queries:
- "Where do we handle authentication?"
- "Find the database connection logic"
- "Show me error handling code"
- "Where is the user validation implemented?"
CodeWeaver redirects all storage to the plugin's designated data directory via XDG environment variables set in the plugin configuration:
- Index data:
${CLAUDE_PLUGIN_DATA}/state/codeweaver/indexes/ - Vector store:
${CLAUDE_PLUGIN_DATA}/state/codeweaver/vectors/ - Model cache:
${CLAUDE_PLUGIN_DATA}/cache/codeweaver/ - Configuration:
${CLAUDE_PLUGIN_DATA}/config/codeweaver/
This ensures data persists across plugin updates and is stored in the standard plugin data directory.
Note: XDG directory overrides apply on Linux. On macOS, storage falls back to
~/Library/...directories unless you configure paths explicitly viaCODEWEAVER_INDEX_STORAGE_PATHand related settings.
For advanced configuration (custom embedding providers, vector stores, etc.), see the main CodeWeaver documentation.
Default configuration uses:
- Embedding Provider: Voyage AI (requires API key) with automatic fallback to local FastEmbed
- Vector Store: Qdrant (in-memory mode, no Docker required)
- Chunking: Language-aware with semantic boundaries
- Python 3.12 or later
- uv package manager (must be installed separately; see the installation guide)
- Documentation: https://github.com/knitli/codeweaver
- Issues: https://github.com/knitli/codeweaver/issues
- Discord: Coming soon
CodeWeaver respects your privacy:
- Code stays local by default (optional cloud embeddings)
- Telemetry is privacy-focused and opt-out
- See Privacy Policy
MIT OR Apache-2.0
Built with ❤️ by Knitli