MCP server and local proxy that compresses LLM prompts, tool output, and replies to cut token cost, with a quality gate that reverts any step that does not save. Exposes llmtrim_compress, llmtrim_compress_text, and llmtrim_stats.
View & run on Rokha →