AI coding tools send raw, uncompressed context to LLMs — full file reads, verbose CLI output, repeated content. That's where your tokens go. LeanCTX intercepts everything between your AI tool and the LLM. It compresses file reads (10 modes, re-reads cost ~13 tokens), shell output (90+ patterns), and session context — automatically. 42 MCP tools. Works with Cursor, Claude Code, Copilot, Windsurf & 20+ more. Single Rust binary. Installs in 60 seconds. No account. No cloud. No telemetry. Ever.