Meta Prompter

A prompt evaluation tool available as both an MCP server and a standalone CLI.

What it does

Evaluates a prompt across 8 dimensions (clarity, specificity, context, actionability, safety, testability, hallucination prevention, token efficiency), computes a weighted global score, and—if needed—returns a full rewrite. Zero fluff, machine‑friendly.

Available as an MCP server and a CLI — both can be used by Claude Code Skills for agentic prompt evaluation workflows.

Quick theory

• On the basis of the paper Meta-Prompting

• The model’s task is to analyze another prompt, not perform the end‑task — prompts about prompts = meta.

• LLM‑as‑a‑Judge: role + rubric + constrained JSON output approximates structured human evaluation.

• Form‑filling graders (e.g., G‑Eval): schema‑locked fields and brief justifications reduce variance and make results comparable.

• Rule‑guided critique (Constitutional‑style): conflict rules, edge‑case handling, and safety checks function as a small “constitution.”

• Reflection/self‑critique patterns: institutionalizes a critique step that boosts reliability before you ship a prompt.

One‑liner: it treats prompts as artifacts to be graded with rules, not instructions to be executed.

Why a separate evaluation model?

Claude Code doesn't allow customizing the LLM temperature. Its default temperature is 1. For more stable scoring, use a lower temperature — this reduces randomness in sampling, ensuring that scores and justifications remain consistent across runs. The evaluation model uses temperature 0.

Both the MCP server and CLI share the same core evaluation logic, so results are identical regardless of how you invoke it.

MCP Server

Use the MCP server for agentic integration — it provides structured output (outputSchema + structuredContent) so agents can act on evaluation data programmatically.

Setup

claude mcp add meta-prompter \
  --env PROMPT_EVAL_MODEL=anthropic:claude-sonnet-4-20250514 \
  --env PROMPT_EVAL_API_KEY=<claude_api_key> \
  -- npx -y meta-prompter-mcp@latest

JSON Configuration

    "meta-prompter": {
      "command": "npx",
      "args": [
        "-y",
        "meta-prompter-mcp@latest"
      ],
      "env": {
        "PROMPT_EVAL_MODEL": "anthropic:claude-sonnet-4-20250514",
        "PROMPT_EVAL_API_KEY": "sk-123456789",
      }
    }

Available Tools

• evaluate - Evaluate a prompt using AI analysis.
  • prompt (string, required): The prompt to evaluate
• ping - Simple ping test to verify connection

CLI

Use the CLI for quick evaluations from the terminal, scripts, or CI pipelines.

# Run directly with npx
npx meta-prompter-mcp "Your prompt here"

# Or install globally
npm install -g meta-prompter-mcp@latest
meta-prompter "Your prompt here"

Usage

meta-prompter [options] [prompt]

Arguments:

• prompt — The prompt to evaluate (or pipe via stdin)

Options:

• --prompt <text> — The prompt to evaluate (alternative to positional arg)
• --model <key> — Model key (default: PROMPT_EVAL_MODEL env or anthropic:claude-sonnet-4-20250514)
• --api-key <key> — API key (default: PROMPT_EVAL_API_KEY env)
• --compact — Output compact JSON instead of pretty-printed
• -h, --help — Show help message
• -v, --version — Show version number

Examples

# Positional argument
PROMPT_EVAL_API_KEY=sk-... meta-prompter "Write a function that sorts an array"

# Flag
meta-prompter --api-key sk-... --prompt "Write a function that sorts an array"

# Pipe from stdin
echo "Write a function that sorts an array" | meta-prompter --api-key sk-...

# Pipe from file
cat my-prompt.txt | meta-prompter --api-key sk-...

# Compact output for piping to jq
meta-prompter --api-key sk-... --compact "test prompt" | jq '.scores.global'

Environment Variables

Both MCP and CLI use the same environment variables:

• PROMPT_EVAL_MODEL - Model key in provider:model-id format (default: anthropic:claude-sonnet-4-20250514, also supports openai:gpt-5)
• PROMPT_EVAL_API_KEY - API key for the chosen provider

The CLI also accepts --model and --api-key flags to override these.

View Eval Results

Each evaluation is appended to evaluation_result.jsonl in the current working directory.

eval-viewer.html is the SPA used to view the eval result jsonl data. Open it in a browser and upload the file.

Testing

# Test with MCP Inspector
npx @modelcontextprotocol/inspector node dist/index.js

Publishing

• Ensure the latest version is committed in the package.json following Semantic Versioning. e.g. "version": "1.1.1"
• Run ./setup.sh
• Run npm publish --tag latest