Promptic — Analisador de Qualidade de Prompts
cppis/promptic
Analisa e melhora a qualidade de seus prompts com feedback detalhado
Ver no GitHubMCP 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/promptic1 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
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
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
- Node.js 18+ — Download
- Claude Desktop or Claude Code — Download Claude Desktop
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
| OS | Config 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
| # | Method | Example |
|---|---|---|
| 1 | Command >> anz / >> 분석 | Build a login form in React >> anz |
| 2 | Command >> anz+run / >> 분석+실행 | Design a REST API >> anz+run |
| 3 | Command >> deep / >> 정밀분석 | Design a system >> deep |
| 4 | Command >> re / >> v2 | >> re (set previous entryId as parentId) |
| 5 | Command >> last N | >> last 2 (reference 2nd most recent entry, N: 1–10) |
| 6 | English natural language | analyze this prompt, how good is this, review this prompt |
| 7 | Korean natural language | 이 프롬프트 분석해줘, 이 프롬프트 어때?, 프롬프트 좀 봐줘 |
| 8 | Implicit improvement request | what'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:
| Command | Behavior |
|---|---|
>> anz / >> 분석 | Returns score, missing elements, and an enhanced prompt block (ready to copy) |
>> anz+run / >> 분석+실행 | Analyzes, then immediately executes the improved prompt |
>> deep / >> 정밀분석 | Full 4-step pipeline: analyze → improve → re-analyze → compare |
>> re / >> v2 | Sets the previous entryId as parentId for version chaining |
>> last N | Sets 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):
| Command | Token increase | Reason |
|---|---|---|
>> 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 mode | Separate from Claude subscription | Direct Anthropic API call, billed independently |
Relative cost by model (same token count):
| Model | Relative cost | Recommended for |
|---|---|---|
| Claude Haiku 4.5 | 1× (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>> deepor Opus for critical prompts. Running>> deepwith 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:
| Preset | Description | Arguments |
|---|---|---|
quick-analyze | Fast 5-axis analysis + enhanced prompt block | prompt (required) |
deep-analyze | API-mode deep analysis (API key required) | prompt (required) |
project-report | Project statistics + dashboard generation | projectId (optional) |
Domain-specific:
| Preset | Description | Arguments |
|---|---|---|
code-review | Build a code review prompt + analyze quality | language (required), context (optional) |
doc-writing | Build a technical doc prompt + analyze quality | docType (required), audience (optional) |
system-design | Build a system design prompt + analyze quality | system (required), constraints (optional) |
Next Steps
| Document | Contents |
|---|---|
| Usage Guide | Tool parameters, usage scenarios, project management, version control, visualization, API mode, FAQ |
| Project Overview | Architecture, roadmap |
Quick Command Reference
| What you want to do | Say 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.