Pqs Mcp Server

Scores prompts across eight dimensions (clarity, specificity, context, constraints, output format, role definition, examples, chain-of-thought structure) before they hit your model. Built on PEEM, RAGAS, MT-Bench, G-Eval, and ROUGE. The free tier gives you 0-80 scores with A-F grades and weakest-dimension feedback, rate-limited per IP. Pro subscribers get optimize_prompt, which rewrites your prompt and runs both versions through a frontier model so you can see before-and-after outputs side by side. Useful as a pre-inference quality gate: block anything below 56/80 to avoid wasting tokens on underspecified requests. Works over stdio or streamable HTTP, with an optional self-hosted backend override.