CodeWeaver

소개

AI coding agents often struggle with context overload, leading to wasted tokens and missed critical patterns due to irrelevant code. CodeWeaver solves this by functioning as an extensible context platform and MCP server that distills and delivers targeted, token-efficient context. It simplifies agent interaction with a single natural language `find_code` tool, eliminating confusion over multiple search methods. By utilizing agent-driven curation and a hybrid search pipeline combining text, semantic embeddings, and AST analysis, CodeWeaver significantly reduces context bloat by 60-80% while enhancing search precision, ensuring agents receive only the most relevant code with exact line and column references.

주요 기능

• Single natural language `find_code` interface for simplified agent interaction.
• 4 GitHub stars
• Agent-driven context curation via MCP sampling to enhance precision without polluting context.
• Pinpoint accuracy with exact line and column references, eliminating 'nearby code' confusion.
• Hybrid search pipeline combining text, semantic embeddings, and AST-aware analysis for comprehensive results.
• Extensible plugin architecture supporting 10+ embedding providers, multiple vector stores, and diverse programming languages (26 full AST, 170+ heuristic).

사용 사례

• Optimizing AI agent performance by delivering precise, relevant code context for development tasks.
• Reducing AI token costs in agent-driven development workflows by minimizing irrelevant context.
• Accelerating developer understanding of unfamiliar or legacy codebases through intelligent search and context delivery.