Patronus AI MCP Server
By thumbsupordown
The Patronus MCP server exposes its LLM evaluation stack to any MCP-compatible application, letting an agent run optimizations, evaluations and experiments against the Patronus SDK. It includes purpose-built judge models — Glider and Lynx — for scoring outputs, end-to-end traces for debugging agent workflows, and the Percival debugger, which looks for failure modes across reasoning, planning and tool execution rather than just grading a final answer.