Codeweaver
Give coding agents hybrid semantic plus traditional code search across a repo with AST-aware chunking for 166 languages.
Overview
com.knitli/codeweaver is a MCP server for the Build phase that provides hybrid AST-aware semantic code search across 166 languages for AI coding agents.
What is this MCP server?
- Hybrid semantic, sparse, and traditional search with reranking
- AST-aware semantic chunking and semantically-aware delimiters
- 166 languages supported per registry metadata
- Many embedding providers (OpenAI, Ollama, Voyage, Bedrock, and others)
- FastMCP-based server; Python >=3.12; uv package manager
- 166 languages_supported per registry capabilities
- 3 search_types: semantic, hybrid, traditional
- 2 chunking_strategies: semantic, semantically-aware-delimiters
Community signal: 11 GitHub stars.
What problem does it solve?
Agents waste tokens skimming whole files because keyword search misses concepts and naive embeddings ignore code structure.
Who is it for?
Solo builders with large or polyglot repos who live in Claude Code and need reliable “find implementation” tooling.
Skip if: Tiny single-file scripts where ripgrep is enough, or teams that refuse to configure embeddings or a vector store.
What do I get? / Deliverables
After indexing and MCP registration, the agent retrieves structurally relevant chunks via hybrid search instead of fragile full-repo grep.
- MCP tools returning ranked code chunks from hybrid search
- AST-aware indexed coverage for supported languages in the project
- Agent-ready navigation across semantic and traditional query modes
Recommended MCP Servers
Journey fit
Agent-facing code understanding is core Build work, so agent-tooling is the primary shelf even when search supports review later. Codeweaver is built for AI agents navigating codebases—not a generic IDE plugin—so it belongs under agent-tooling.
How it compares
MCP code-search appliance with AST chunking, not a markdown planning skill or generic web MCP.
Common Questions / FAQ
Who is com.knitli/codeweaver for?
Developers running AI agents on real codebases who need semantic and hybrid search beyond basic text match.
When should I use com.knitli/codeweaver?
Use it during Build and Ship review when the agent must locate definitions, call sites, or patterns across many languages or a large monorepo.
How do I add com.knitli/codeweaver to my agent?
Run the FastMCP Codeweaver server (Python >=3.12, typically via uv), point it at your repo and embedding provider, then register the stdio or local MCP URL in Claude Code or Cursor.