LeanCTX acts as a cognitive context layer between code and AI models, specifically targeting AI coding assistants like Claude Code, Cursor, and Copilot. It compresses file reads and git status outputs, caches context across sessions to avoid redundant token usage, and provides 63 MCP tools for context routing. The system persists memory across chats and includes a dashboard for token budget monitoring, aiming to reduce context overhead by up to 99%.