Token Compressor

Token Compressor is an open-source MCP server and Python toolkit for compressing language model prompts by 30–70% while preserving semantic meaning and conditional logic. Using a two-stage pipeline, it rewrites prompts to their semantic minimum and validates meaning retention with embedding cosine similarity, which helps reduce token costs and optimize LLM workflows. It is ideal for LLM developers, prompt engineers, and teams seeking more efficient, cost-effective context management when working with local or API-based models.