Compress tool outputs, logs, JSON, files, and RAG chunks with Headroom (local-first library, proxy, and MCP) before they reach the LLM. Same answers, fewer tokens; originals stay retrievable via CCR. Use when the user mentions Headroom, headroom-ai, token compression, context bloat, wrapping Cursor/Claude/Codex, MCP tools headroom_compress / headroom_retrieve / headroom_stats, or when large tool dumps threaten the context window and Headroom MCP is available.
