Pqs Mcp Server
Score and optionally optimize agent prompts on eight dimensions across five frameworks before you burn tokens on a weak run.
Overview
PQS MCP is a Build-phase MCP server that scores and can optimize prompts across eight dimensions and five frameworks before your model runs.
What is this MCP server?
- Free score_prompt tool with no API key; Pro optimize_prompt via PQS_API_KEY
- Eight scoring dimensions and five evaluation frameworks in one pre-run pass
- Streamable HTTP remote at promptqualityscore.com/api/mcp plus npm stdio package
- Pre-flight gate: fix clarity and structure before the model executes
- Version 1.4.0 in the official MCP registry schema
- 8 scoring dimensions
- 5 evaluation frameworks
- Server version 1.4.0
Community signal: 2 GitHub stars.
What problem does it solve?
You only discover a sloppy or ambiguous prompt after wasted tokens and confusing agent behavior.
Who is it for?
Indie builders who iterate Claude/Cursor agent prompts daily and want a quick numeric rubric without building their own eval harness.
Skip if: Teams that only need post-response grading or offline benchmark suites with no MCP client.
What do I get? / Deliverables
You get a structured quality score (and optional optimized text) before execution so agent runs start from a stronger instruction set.
- Dimensional PQS score for a draft prompt
- Optional optimized prompt text via Pro API
- Repeatable pre-run quality gate in your agent loop
Recommended MCP Servers
Journey fit
How it compares
MCP pre-flight prompt scorer, not a general coding skill or output log analyzer.
Common Questions / FAQ
Who is PQS MCP for?
Solo and small-team builders using MCP-enabled agents who want structured prompt QA before each run.
When should I use PQS MCP?
Before sending important agent tasks—new features, refactors, or customer-facing automations—when prompt drift costs time.
How do I add PQS MCP to my agent?
Register the streamable-http remote URL or install the npm pqs-mcp-server stdio package; set PQS_API_KEY only if you use optimize_prompt.