"author:ZettaQuant" matching MCP servers:

Enables any MCP-aware LLM host to filter noisy text down to query-relevant sentences via a topic-conditioned relevancy classifier, reducing token spend before the model processes the full input.