Prompt-Optimizer

Prompt-Optimizer (PCP Engine) is an MCP server that analyzes, scores, enforces policy, and provides a guided workflow for refining prompts. It can classify prompts, assess risk, route to appropriate models, and prepare a compiled, policy-compliant prompt along with a structured audit trail. The built-in commands cover a full cycle from quick quality checks to an end-to-end optimization pipeline. Typical usage includes running a preflight to understand risks, then using optimize to produce a surfaced PreviewPack for approval, and finally enforcing the finalized prompt through the chosen model route. This makes it suitable for teams that want deterministic prompt quality and governance without leaking into the LLMs themselves.

To use the available tools, install the PCP Engine CLI (pcp-engine) and run commands such as preflight, optimize, check, score, cost, and benchmark. Preflight classifies the prompt, scores 5 dimensions, and surfaces blocking questions if needed. Optimize executes the full pipeline: analyze, compile, surface blocking questions, and produce a PreviewPack for approvals. The check command provides a quick quality score and top issues, while score gives a detailed 5-dimension breakdown. Cost estimates token usage across supported models, and benchmark runs regression tests to ensure stability. The workflow supports human-in-the-loop approvals, where blocking questions must be answered before finalizing the compiled prompt.

In CI or GitHub Actions, you can wire PCP to run preflight or optimize as part of your quality gates, ensuring prompts meet your governance criteria before deployment or release.

Prerequisites:

• Node.js 18+ and npm installed on your system
• Access to npm to install the PCP Engine CLI

Installation steps:

1. Install the PCP Engine CLI globally (recommended for quick start):

   npm install -g pcp-engine

2. Verify installation:

   pcp --version pcp preflight "Your prompt here" --json

3. Optional: if you prefer to run without global installation, you can use npx directly (as in the example below):

   npx -y pcp-engine preflight "Your prompt here" --json

4. Review and customize environment variables as needed (see additional_notes for details):

• PCP_LOG_LEVEL: controls verbosity (debug|info|warn|error)
• PCP_FREE_TIER: enable/disable free tier features for testing

Tips and common considerations:

• The CLI commands mirror the MCP workflow: preflight, optimize, check, score, cost, benchmark. Use preflight for quick risk assessment and optimization for full pipeline results.
• The free tier typically provides limited optimizations per month; monitor usage if you’re evaluating scale.
• If running in CI, consider using the --json flag to get structured output for parsing in scripts.
• Environment variables can tune behavior (logging, tier limits). In production, securely manage any sensitive values the MCP may reference.
• When integrating with GitHub Actions or CI, pin the PCP Engine version to avoid unexpected changes, and ensure actions/checkout is configured for file access during prompts processing.