ui-extractor
MCP server for extracting UI components, design tokens, and patterns from websites and generating production-ready React, Vue, and HTML components.
README
UI Extractor
Extract UI components, design tokens, and patterns from any website. Generate production-ready React, Vue, and HTML components. Includes CLI and MCP server for AI agent integration.
Table of Contents
- Features
- Tech Stack
- Getting Started
- Architecture
- Project Structure
- Scripts
- Configuration
- Agent Rules
- License
Features
- Extract — DOM, CSS variables, computed styles, images, screenshots (full page, viewport, mobile, sections, components)
- Analyze — Pattern detection, design system extraction (colors, spacing, radius, fonts), component classification
- Spec — Generate component specifications (props, states, accessibility, responsive breakpoints)
- Generate — Output HTML, React, or Vue components with CSS/SCSS/Tailwind
- Pipeline — Full extract → analyze → spec → generate in one command
- Learn — Save any site as a reference for future regeneration
- MCP Server — Integrate with Cursor, Claude, Windsurf, VS Code
Tech Stack
- Language — TypeScript (ESM)
- Runtime — Node.js 20+
- Package Manager — pnpm 9+
- Browser Automation — Playwright
- CLI Framework — Commander.js
- MCP SDK — @modelcontextprotocol/sdk
- Testing — Vitest
- Linting — ESLat 9+ with custom rules (STD-DOC-002, STD-DOC-003)
Getting Started
Prerequisites
- Node.js 20 or higher
- pnpm 9 or higher
Installation
git clone https://github.com/stsgs1980/ui-extractor.git
cd ui-extractor
pnpm install
Build
pnpm build
Run CLI
# Full pipeline: extract + analyze + spec + generate
pnpm --filter @ui-extractor/cli start pipeline https://ui.shadcn.com --component button --format react
# Extract only
pnpm --filter @ui-extractor/cli start extract https://ui.shadcn.com --screenshot
# Analyze only
pnpm --filter @ui-extractor/cli start analyze https://ui.shadcn.com
# Generate from URL
pnpm --filter @ui-extractor/cli start generate https://ui.shadcn.com --component button --format react
# Learn a site for later use
pnpm --filter @ui-extractor/cli start learn https://ui.shadcn.com --save shadcn-ui
# List saved references
pnpm --filter @ui-extractor/cli start references
# Generate from saved reference
pnpm --filter @ui-extractor/cli start generate-ref shadcn-ui --format vue
# Screenshots
pnpm --filter @ui-extractor/cli start screenshot https://ui.shadcn.com --types full,viewport,sections,components
# Compare two sites
pnpm --filter @ui-extractor/cli start compare https://ui.shadcn.com https://vercel.com
Run MCP Server
Add to your MCP config (Cursor, Claude Desktop, Windsurf, VS Code):
{
"mcpServers": {
"ui-extractor": {
"command": "node",
"args": ["packages/mcp/src/server.js"],
"cwd": "/path/to/ui-extractor"
}
}
}
Then in Cursor:
@ui-extractor uix_pipeline https://ui.shadcn.com --component button --format react
Architecture
Monorepo with three packages:
packages/core— Core library (extract, analyze, spec, generate, pipeline, learn)packages/cli— Command-line interfacepackages/mcp— MCP server for AI agent integration
Data flow: URL → Playwright extraction → Analysis (design system, components, patterns) → Spec generation → Code generation (React/Vue/HTML).
Project Structure
packages/core/src/— Core modules (extract.js, analyze.js, spec.js, generate.js, pipeline.js, index.js)packages/cli/src/— CLI entry point (index.js)packages/mcp/src/— MCP server (server.js)docs/— Documentation (agent-workflow.md, teardown-method.md, teardown-cheatsheet.md)eslint-rules/— Custom ESLint rules (unicode-policy, code-block-language)eslint-processors/— Custom ESLint processors (markdown-snippets)
Scripts
| Script | Description |
|---|---|
pnpm dev |
Run CLI in watch mode |
pnpm build |
Build all packages |
pnpm test |
Run tests with Vitest |
pnpm eslint . |
Lint all files |
pnpm mcp |
Start MCP server |
pnpm cli |
Run CLI |
Configuration
Screenshot Types
full— Full page screenshotviewport— Viewport onlymobile— Mobile viewport (375px)sections— Per-section screenshotscomponents— Per-component screenshots
pnpm --filter @ui-extractor/cli start extract https://example.com --screenshot --screenshot-types full,viewport,sections,components
MCP Tools
uix_pipeline— Full pipelineuix_extract— Extract UI datauix_analyze— Analyze patterns and design systemuix_spec— Generate component specificationuix_generate— Generate component codeuix_learn— Learn and save referenceuix_references— List referencesuix_generate_from_reference— Generate from saved referenceuix_compare— Compare two URLs
Agent Rules
Any AI agent working on this project MUST read and follow AGENT_RULES.md before performing any operations.
License
MIT
Built with: Node.js 20 + TypeScript + Playwright + Commander.js
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。