LexRAG

概要

LexRAG is an MCP-native server designed to provide memory and code intelligence for LLM agents, transforming code repositories into queryable graphs. It addresses common RAG agent limitations by offering persistent memory across sessions, efficient context management, and cross-file reasoning capabilities. This enables agents to answer complex architectural, impact, and planning questions without repeatedly processing the entire codebase, thereby saving token budgets and improving development speed.

主な機能

• Code graph intelligence for cross-file dependency answers
• Impact-scoped test selection for smarter test runs
• 1 GitHub stars
• Persistent agent memory for decisions and episodes across sessions
• Hybrid retrieval (vector + BM25 + graph) for enhanced relevance
• Temporal model for historical queries and change diffs

ユースケース

• Understand a subsystem by querying dependency maps and concrete code slices
• Coordinate multiple agents with ownership, task context, and durable progress tracking
• Plan safe changes by analyzing impact and running focused tests