Prometheus-MCP
Prometheus-MCP is an AI Creative Director MCP server that enhances creative output from models like Claude and GPT to expert-level quality through iterative improvement loops.
README

<div align="center">
Prometheus-MCP
Creative Director MCP
AI Creative Director that turns ordinary model output into expert-level work.
</div>
한국어
Prometheus-MCP는 Claude, GPT, DeepSeek, MiniMax, Qwen, GLM 및 미래 모델이 생성한 크리에이티브 산출물을 전문가 수준으로 끌어올리는 Model Context Protocol 서버입니다. 단순한 문서 검색 시스템이 아니라, 생성 - 평가 - 개선 - 재생성 루프를 통해 품질을 지속적으로 향상시키는 AI Creative Director입니다.
대상 분야
- 웹 디자인 / UI 디자인 / UX 디자인
- Three.js / React Three Fiber / WebGL / WebGPU
- VFX / 셰이더 (GLSL / WGSL) / 인터랙티브 경험
- 프론트엔드 애니메이션 / 크리에이티브 코딩
- 게임 개발 (2D / 3D / Phaser / Rapier / ECS)
핵심 가치
- 전문가 패턴 라이브러리 - 22개 패턴, 확장 가능한 구조
- Critic Engine - 16개 품질 차원, 60개 규칙, 증거 기반 평가
- 자동 개선 루프 - 목표 점수 도달까지 반복
- 품질 인텔리전스 - 감사 가능한 점수 근거
- AI 크리에이티브 디렉팅 - 패턴 선택, 개선 전략, 재생성
아키텍처
사용자 요청
-> Planner
-> Knowledge Collector
-> Pattern Selector
-> Prompt Enhancer
-> Generation Provider
-> Evidence Collector
-> Critic Engine
-> Quality Scoring
-> Improvement Engine
-> Regeneration Loop
-> 최종 결과
모듈
| 모듈 | 역할 |
|---|---|
| planner | 도메인 탐지, 브리프 생성, 종료 정책 |
| knowledge | 외부 지식 수집 + 프롬프트 인젝션 방어 |
| patterns | 패턴 검증, 저장소, 가중치 기반 선택 |
| critic | 증거 수집(결정적 측정) + 규칙 평가 + 점수 + 감사 가능 근거 |
| improver | 부분/전체 재생성 전략, 수정 프롬프트 |
| loop_controller | 상태 머신 + 종료 정책(목표/최대반복/비용/시간/수익체감) |
| providers | 역량 기반 라우터 (Stub, OpenAI 호환, Claude) |
| memory | 세션 간 학습, 효과성 추적 |
| history | 세션 내 타임라인 |
| telemetry | 구조화 로그, 메트릭, 비용, 트레이싱, 비밀 마스킹 |
| infrastructure | 설정, 캐시, 보안 |
| mcp | 7 tools / 5 resources / 4 prompts |
MCP 도구
| 도구 | 설명 |
|---|---|
| direct_creative_work | 브리프에서 생성-평가-개선 루프 실행 |
| critique_artifact | 산출물 평가 (점수, 강점, 약점, 제안) |
| improve_artifact | 평가 기반 개선 계획 + 수정 프롬프트 |
| list_patterns | 패턴 목록 조회 |
| get_pattern | 패턴 상세 조회 |
| recall_sessions | 과거 세션 조회 |
| get_quality_trends | 품질 추세 분석 |
설치 및 실행
npm install
npm run build
npm start
opencode 로컬 MCP 연결
프로젝트 루트에 opencode.json을 생성하고 Prometheus를 로컬 MCP로 등록합니다. API 키 불필요 (StubProvider 기본 활성화).
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"prometheus": {
"type": "local",
"command": ["node", "E:/Downloads/Prometheus-MCP/dist/index.js"],
"enabled": true
}
}
}
opencode 재시작 후 7개 도구(direct_creative_work, critique_artifact 등)가 에이전트에 연결됩니다.
Claude Desktop 연결
claude_desktop_config.json에 추가:
{
"mcpServers": {
"prometheus": {
"command": "node",
"args": ["E:/Downloads/Prometheus-MCP/dist/index.js"]
}
}
}
테스트
npm test
설정
환경 변수로 런타임 설정을 덮어쓸 수 있습니다.
| 변수 | 설명 |
|---|---|
| PROMETHEUS_TARGET_SCORE | 목표 품질 점수 (기본 85) |
| PROMETHEUS_MAX_ITERATIONS | 최대 개선 반복 (기본 5) |
| PROMETHEUS_MAX_COST_USD | 비용 상한 (USD) |
| PROMETHEUS_DEFAULT_PROVIDER | 기본 모델 제공자 |
| PROMETHEUS_ENABLE_LLM_REASONING | LLM 추론 평가 활성화 |
| PROMETHEUS_ALLOW_COMMUNITY_PATTERNS | 커뮤니티 패턴 허용 |
모델 제공자 API 키는 해당 환경 변수(OPENAI_API_KEY, ANTHROPIC_API_KEY, DEEPSEEK_API_KEY 등)가 존재하면 자동 활성화됩니다.
기술 스택
- TypeScript (ES Modules)
- Node.js 20+
- @modelcontextprotocol/sdk 1.29.0 (안정版)
- zod v3 / ajv
- Vitest
라이선스
MIT
English
Prometheus-MCP is a Model Context Protocol server that elevates creative output from Claude, GPT, DeepSeek, MiniMax, Qwen, GLM, and future models to expert-level quality. It is not a document retrieval system. It is an AI Creative Director that runs a generate - critique - improve - regenerate loop to continuously raise quality.
Target Domains
- Web design / UI design / UX design
- Three.js / React Three Fiber / WebGL / WebGPU
- VFX / shaders (GLSL / WGSL) / interactive experiences
- Frontend animation / creative coding
- Game development (2D / 3D / Phaser / Rapier / ECS)
Core Value
- Expert Pattern Library - 22 patterns, extensible structure
- Critic Engine - 16 quality dimensions, 60 rules, evidence-bound evaluation
- Auto Improvement Loop - iterates until target score is reached
- Quality Intelligence - auditable score justifications
- AI Creative Directing - pattern selection, improvement strategy, regeneration
Architecture
User request
-> Planner
-> Knowledge Collector
-> Pattern Selector
-> Prompt Enhancer
-> Generation Provider
-> Evidence Collector
-> Critic Engine
-> Quality Scoring
-> Improvement Engine
-> Regeneration Loop
-> Final result
Modules
| Module | Responsibility |
|---|---|
| planner | domain detection, brief formation, termination policy |
| knowledge | external knowledge collection + prompt injection defense |
| patterns | pattern validation, repository, weighted selection |
| critic | evidence collection (deterministic measurement) + rule evaluation + scoring + auditable justification |
| improver | surgical/full regeneration strategy, revision prompts |
| loop_controller | state machine + termination policy (target/maxIter/cost/wallClock/diminishing returns) |
| providers | capability-based router (Stub, OpenAI-compatible, Claude) |
| memory | cross-session learning, effectiveness tracking |
| history | intra-session timeline |
| telemetry | structured logs, metrics, cost, tracing, secret redaction |
| infrastructure | config, cache, security |
| mcp | 7 tools / 5 resources / 4 prompts |
MCP Tools
| Tool | Description |
|---|---|
| direct_creative_work | runs generate-critique-improve loop from a brief |
| critique_artifact | evaluates an artifact (score, strengths, weaknesses, suggestions) |
| improve_artifact | builds improvement plan + revision prompt from a critique |
| list_patterns | lists available patterns |
| get_pattern | gets pattern detail |
| recall_sessions | recalls past sessions |
| get_quality_trends | analyzes quality trends |
Install and Run
npm install
npm run build
npm start
opencode Local MCP
Create opencode.json in your project root and register Prometheus as a local MCP. No API key needed (StubProvider enabled by default).
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"prometheus": {
"type": "local",
"command": ["node", "E:/Downloads/Prometheus-MCP/dist/index.js"],
"enabled": true
}
}
}
Restart opencode and the 7 tools (direct_creative_work, critique_artifact, etc.) become available to the agent.
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"prometheus": {
"command": "node",
"args": ["E:/Downloads/Prometheus-MCP/dist/index.js"]
}
}
}
Tests
npm test
Configuration
Environment variables override runtime config.
| Variable | Description |
|---|---|
| PROMETHEUS_TARGET_SCORE | target quality score (default 85) |
| PROMETHEUS_MAX_ITERATIONS | max improvement iterations (default 5) |
| PROMETHEUS_MAX_COST_USD | cost cap (USD) |
| PROMETHEUS_DEFAULT_PROVIDER | default model provider |
| PROMETHEUS_ENABLE_LLM_REASONING | enable LLM reasoning in critique |
| PROMETHEUS_ALLOW_COMMUNITY_PATTERNS | allow community-trust patterns |
Model provider API keys auto-enable their providers when the corresponding env var (OPENAI_API_KEY, ANTHROPIC_API_KEY, DEEPSEEK_API_KEY, etc.) is present.
Tech Stack
- TypeScript (ES Modules)
- Node.js 20+
- @modelcontextprotocol/sdk 1.29.0 (stable)
- zod v3 / ajv
- Vitest
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.