LeanCTX

About

AI coding tools transmit raw, uncompressed context to large language models, including full file reads, lengthy CLI outputs, and repeated content. This is what consumes your tokens. LeanCTX positions itself between your AI tool and the LLM to intercept all such data. It automatically compresses file reads with ten modes, where re-reads cost approximately thirteen tokens, and condenses shell output across over ninety patterns, along with session context. It includes forty two MCP tools, integrates with Cursor, Claude Code, Copilot, Windsurf, and twenty plus others, and runs as a single Rust binary. Installation takes sixty seconds. It requires no account, uses no cloud services, and collects no telemetry, ever.