MCP server that compresses LLM responses into caveman-style prose, reducing token usage by ~65% while maintaining technical accuracy.
