convez.ai
MCP ServerProdutividadeIniciante
cppis

Promptic — Analisador de Qualidade de Prompts

cppis/promptic

Analisa e melhora a qualidade de seus prompts com feedback detalhado

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que analisa prompts, pontuando sua qualidade e revelando o que a IA compreendeu, o que inferiu e o que deixou passar despercebido. Rastreia melhorias ao longo do tempo e funciona nativamente com Claude Desktop e Claude Code, sem necessidade de API key.

Instalação

$npx convez-ai cppis/promptic

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 7 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Refinamento iterativo de prompts para IA
  • Auditoria de compreensão de instruções
  • Rastreamento de melhoria em engenharia de prompts

Veredito

Ferramenta prática para quem trabalha intensamente com prompts no Claude. O feedback sobre o que a IA inferiu vs. perdeu é valioso, mas a utilidade depende de usar regularmente e incorporar os aprendizados — não é mágica, é espelho.

README

A tradução para o português deste README ainda está na fila.

Promptic — Getting Started

EN KO

npm version npm downloads Node.js License: MIT MCP Compatible GitHub release GitHub last commit GitHub issues GitHub stars

Promptic is an MCP (Model Context Protocol) server for Claude Desktop and Claude Code that makes your prompts visible — what the AI understood, what it guessed, and what it couldn't know.

Trigger with >> anz, >> 분석, or just say "analyze this prompt." No API key required for local mode.

Get Promptic installed and analyze your first prompt in 5 minutes.


Prerequisites


Step 1. Register the MCP Server

Option A: Run from source (for development)

git clone https://github.com/cppis/prompt-lens.git
cd prompt-lens/mcp-server
npm install

Run the setup script to register Promptic with Claude Desktop automatically:

./scripts/setup-claude-desktop.sh

The script detects your OS and adds Promptic to claude_desktop_config.json without affecting existing MCP configurations.

For Claude Code:

claude mcp add promptic node /your/path/to/prompt-lens/mcp-server/index.js
OSConfig file path
macOS~/Library/Application Support/Claude/claude_desktop_config.json
Windows%APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "promptic": {
      "command": "node",
      "args": ["/your/path/to/prompt-lens/mcp-server/index.js"]
    }
  }
}

Replace /your/path/to/ with the actual path where you cloned the repo. Restart Claude Desktop after saving.

Option B: Run via npx (after npm release)

Once published to npm, no cloning is required.

{
  "mcpServers": {
    "promptic": {
      "command": "npx",
      "args": ["-y", "@cppis/promptic"]
    }
  }
}

Claude Code:

claude mcp add promptic -- npx -y promptic

Step 2. Restart Claude Desktop

After saving the config file, fully quit and relaunch Claude Desktop.

When registered successfully, the MCP tool icon (🔧) appears in the bottom-right of the input area and promptic is listed in the tools menu.

Verify in Claude Code:

claude mcp list

If promptic appears in the output, you're good to go.


Step 3. Analyze Your First Prompt

Promptic is triggered automatically when it detects any of the following:

Trigger Rules

#MethodExample
1Command >> anz / >> 분석Build a login form in React >> anz
2Command >> anz+run / >> 분석+실행Design a REST API >> anz+run
3Command >> deep / >> 정밀분석Design a system >> deep
4Command >> re / >> v2>> re (set previous entryId as parentId)
5Command >> last N>> last 2 (reference 2nd most recent entry, N: 1–10)
6English natural languageanalyze this prompt, how good is this, review this prompt
7Korean natural language이 프롬프트 분석해줘, 이 프롬프트 어때?, 프롬프트 좀 봐줘
8Implicit improvement requestwhat's wrong with this, how can I improve this

For rules #1–3, append the command at the end of your message. Everything before it is treated as the prompt to analyze.

Command behavior comparison:

CommandBehavior
>> anz / >> 분석Returns score, missing elements, and an enhanced prompt block (ready to copy)
>> anz+run / >> 분석+실행Analyzes, then immediately executes the improved prompt
>> deep / >> 정밀분석Full 4-step pipeline: analyze → improve → re-analyze → compare
>> re / >> v2Sets the previous entryId as parentId for version chaining
>> last NSets the Nth most recent entryId as parentId (N: 1–10, no entryId lookup needed)

Token Usage Guide

Token overhead varies by command. Model choice is the biggest factor.

Token increase per command (vs. plain conversation, Sonnet baseline):

CommandToken increaseReason
>> anz (local)+20–30%~500-token JSON result added to context. No LLM call for the analysis itself.
>> anz+run+50–80%Analysis JSON + actual task execution with the improved prompt
>> deep / >> 정밀분석+200–300%2× analyze_prompt calls + improved prompt generation + comparison, 4 steps total
api modeSeparate from Claude subscriptionDirect Anthropic API call, billed independently

Relative cost by model (same token count):

ModelRelative costRecommended for
Claude Haiku 4.51× (baseline)Fast iterative analysis
Claude Sonnet 4.6~4×General use (recommended)
Claude Opus 4.6~20×High-stakes prompts only

Tip: For day-to-day analysis use >> anz + Sonnet, and reserve >> deep or Opus for critical prompts. Running >> deep with Opus can consume up to 10–15× more tokens than a plain conversation.


Example — command trigger:

Build a login form in React >> anz

Example — natural language trigger:

Analyze this prompt: "Build a login form in React"

Promptic responds automatically:

📊 Overall score: 43/100 (Grade D)

5-axis scores:
  - Clarity:       55
  - Specificity:   30
  - Context:       35
  - Structure:     40
  - Actionability: 50

❌ Missing elements: Role, Context, Output Format, Example, Constraints

💡 Suggestions:
  - Define a role (e.g., "You are a senior React developer")
  - Specify the exact requirements for the login form
  - State the desired output format

✨ Improved prompt:
  "You are a senior React developer. Build a login form component using TypeScript.
   Include email/password inputs, validation, and error messages.
   Style it with Tailwind CSS and provide usage instructions alongside the code."

MCP Prompts Workflow (Claude Desktop / menu)

Type / in the Claude Desktop chat input to access Promptic workflow presets:

General:

PresetDescriptionArguments
quick-analyzeFast 5-axis analysis + enhanced prompt blockprompt (required)
deep-analyzeAPI-mode deep analysis (API key required)prompt (required)
project-reportProject statistics + dashboard generationprojectId (optional)

Domain-specific:

PresetDescriptionArguments
code-reviewBuild a code review prompt + analyze qualitylanguage (required), context (optional)
doc-writingBuild a technical doc prompt + analyze qualitydocType (required), audience (optional)
system-designBuild a system design prompt + analyze qualitysystem (required), constraints (optional)

Next Steps

DocumentContents
Usage GuideTool parameters, usage scenarios, project management, version control, visualization, API mode, FAQ
Project OverviewArchitecture, roadmap

Quick Command Reference

What you want to doSay this to Claude
Analyze a prompt (quick)... >> anz or ... >> 분석
Analyze + run immediately... >> anz+run or ... >> 분석+실행
Deep analysis (4-step)... >> deep or ... >> 정밀분석
Analyze via natural language"Analyze this prompt: ..."
Re-analyze previous prompt>> re or >> v2
Reference Nth recent entry>> last N (e.g. >> last 2)
Create a project"Create a project called 'my-project' in Promptic"
Set active project"Set 'my-project' as the active project"
Visualize project... >> viz or ... >> 시각화
Compare prompts... >> diff

For the full command list, see the Usage Guide.

Ver o original no GitHub

Bom para

Relacionados