An LLM-powered vulnerability auditor for MCP servers that catches semantically-equivalent attacks by using a local LLM grounded by retrieval against known attack patterns.
