"author:compresh" matching MCP servers:

Provides production-grade context compression for LLM agent conversations with Q-protective ranking, epistemic markers, and semantic store, reducing token usage while preserving equivalence.