Design Systems Mcp

MCP设计创作

提供方 southleft

设计创作 HTTP · 远程 OAuth 授权 社区 / 第三方 Claude DesktopClaudeCursorVS Code
查看接入方式

I'm your specialized design systems assistant. Ask me about components, tokens, patterns, and best practices.

一句话选型结论

按资料字段整理

Design Systems Mcp 属于设计创作类 MCP 服务器,HTTP · 远程,OAuth 授权,可直接接入 Claude Desktop、Claude、Cursor 等,有可直接复制的远程端点配置。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类设计创作 传输方式HTTP · 远程 鉴权OAuth 授权 部署方式远程部署
GitHub 星标 未标注 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

设计创作
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

mcpai assistant

安装与配置

HTTP · 远程 OAuth 授权
连远程 HTTPS 端点,通常为 Streamable HTTP
https://design-systems-mcp.southleft.com/mcp

把它填进客户端的 MCP 配置即可,不需要在本地安装运行时。

  1. 1
    准备运行环境

    确认你的客户端支持远程端点,并拿到服务地址。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
出图与素材

生成配图、图标与素材。

设计稿处理

读取设计稿、批量改尺寸或导出。

内容生成

按主题生成文案与视觉草稿。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Claude、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • 远程端点依赖网络与服务方可用性,注意密钥与调用额度。
  • 使用前要准备好OAuth 授权对应的凭据,并按最小权限配置。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在设计创作场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Design Systems Mcp当前 未标注 HTTP · 远程 远程部署 有 未标注
Cursor Talk To Figma MCP Da1c7911 40 未标注 未标注 无 2026-05-17
Blender MCP 17 stdio · 本地进程 本地部署 有 2026-05-18
Nova Canvas MCP Server 2 stdio · 本地进程 本地部署 有 2026-05-17
Figma Context MCP 3aaa8d28 未标注 未标注 支持远程部署 无 2026-05-17
kreuzberg 未标注 stdio · 本地进程 本地部署 有 2026-04-12

服务器档案

设计创作
字段来自项目仓库与公开索引,中文说明由本站标注
提供方southleft上游标注的维护方 southleft
分类设计创作本站按上游分类展示
连接方式HTTP · 远程连远程 HTTPS 端点,通常为 Streamable HTTP
鉴权方式OAuth 授权远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · node本地安装时用到的包标识与运行时
部署类型远程部署上游给出的部署形态
最近核验未标注上游最后核验时间

详细介绍

16.7 KB
来自项目仓库与上游目录的原文,未做删改

Design Systems MCP Server

An AI-powered Model Context Protocol (MCP) server providing intelligent access to authoritative design systems knowledge. 395 curated entries spanning W3C standards, WCAG guidelines, design system practice, and — the part a general-purpose model cannot help you with — the 2025-2026 agent-interface protocols.

🌐 Live Demo: https://design-systems-mcp.southleft.com/

Why this exists

A frontier model already knows Atomic Design, BEM, and the WCAG success criteria. It does not reliably know that MCP Apps moved from _meta["openai/outputTemplate"] to _meta.ui.resourceUri, or that A2UI reached v1.0 candidate in June 2026. On fast-moving specs, every model's training is stale in a way it cannot detect — it answers confidently with last year's field names.

That is what this server is for: dated, cited, extracted primary-source content that beats both a smaller model's guess and a frontier model's expensive live re-research.

Features

Core Capabilities

  • 🎯 Vector + Keyword Search - Supabase pgvector with pluggable embeddings (Cloudflare Workers AI bge-m3, edge-native and free within the Workers plan; OpenAI optional), plus a Postgres full-text path that keeps working when embeddings are unavailable
  • 📚 395 Curated Entries - W3C standards, WCAG 2.2, ARIA practices, major design systems, and agent-interface protocols
  • 🚦 Relevance Floor - IDF-weighted scoring means an uncovered topic returns nothing rather than confident, adjacent content
  • 🚀 Edge-Optimized - Cloudflare Workers deployment with global distribution

Latest Updates

  • 🤖 Agent-Interface Coverage (Aug 2026) - A2UI v1.0, MCP Apps SEP-1865, AG-UI, Sentient Design, generative/ephemeral UI, component contracts
  • 🧪 Eval Harness - scripts/eval-mcp.ts scores retrieval and substance across four difficulty tiers
  • 🛡️ Source Reliability Badges - Every answer flags Primary / Authoritative / Reference / Example / Community sources
  • 📖 Universal MCP Client Support - Works with any MCP-capable client (Claude Desktop, Cursor, Windsurf, Codex, etc.)
  • 🏛️ Major Design Systems, Deeply - Carbon, Polaris, Atlassian, Material 3, Fluent, Spectrum, Nord, Mantine, shadcn/ui, Radix, Untitled UI — the guidance and token architecture, not just prop tables

Developer Experience

  • 🌐 Zero Setup Required - Public MCP endpoint ready to use
  • 🤖 AI Chat Interface - Natural-language Q&A grounded in the knowledge base, streamed via Cloudflare Workers AI (Llama 3.3 70B) — no OpenAI, no API credits
  • 🧪 Local Development - Complete testing environment with hot reload
  • 📝 Comprehensive Docs - Updated setup guides for every major MCP client

Content Library

395 Curated Entries Including:

Agent Interfaces & AI (2025-2026 — the material a general model gets wrong)
- A2UI Protocol v1.0 (Google) — adjacency-list components, catalog negotiation, A2A binding
- MCP Apps / SEP-1865 — ui:// resources, CSP metadata, the ui/ method namespace, host theming variables
- AG-UI event reference; OpenAI Apps SDK → MCP Apps field-level migration map
- Sentient Design (Josh Clark) — the framework, the triangle, radically adaptive experiences
- Generative & ephemeral UI — Google Research's generative UI paper, content-vs-chrome
- Component contracts and components-as-data (Nathan Curtis, Christine Vallaure)
- AGENTS.md / SKILL.md / DESIGN.md — the agent-facing design system layer
- How Lovable, v0, Figma Make, and Replit each ingest a design system
- The source-of-truth debate: three camps, and why "code-led vs design-led" isn't the real vocabulary
- Context-Based Design Systems, component contracts, and machine-readable documentation
- How a design system reaches Claude Code, Copilot, Cursor, and Windsurf

Standards & Specifications
- W3C Design Tokens Community Group (DTCG) Specification
- WCAG 2.2 Guidelines (A, AA, AAA levels)
- WAI-ARIA Authoring Practices Guide (APG)
- W3C Web Content Accessibility Guidelines
- W3C Mobile Accessibility at W3C

Design System Resources
- Material Design 3 (Google)
- Fluent Design System (Microsoft)
- Ant Design (Alibaba)
- Carbon Design System (IBM)
- Polaris (Shopify)
- Lightning Design System (Salesforce)
- Atlassian Design System
- Adobe Spectrum
- GitHub Primer
- Shopify Polaris

Tools & Frameworks
- Figma Design System Guides
- Style Dictionary Documentation
- Design Tokens Format Module
- Storybook Best Practices

Methodologies & Best Practices
- Atomic Design principles
- Design Systems Handbook
- Component architecture patterns
- Accessibility implementation guides

Quick Start

Using the Public MCP Server (Recommended)

No installation needed! Connect any MCP client to our live server:```
https://design-systems-mcp.southleft.com/mcp


See Connect to MCP Clients section below for detailed setup instructions.

### Local Development

1. **Clone and Install**```bash
   git clone https://github.com/southleft/design-systems-mcp.git
   cd design-systems-mcp
   npm install
   ```

2. **Configure Environment**```bash
   cp .dev.vars.example .dev.vars
   # Edit .dev.vars and add your credentials
   ```

3. **Start Development Server**```bash
   npm run dev
   ```
   Server available at: `http://localhost:8787`

## Connect to MCP Clients

Choose your AI coding tool below for setup instructions:

Claude Desktop - Click to expand configuration

**Add via Custom Connector UI (Recommended - No JSON editing!)**

1. **Open Claude Desktop** and navigate to **Settings** → **Connectors**

2. **Click "Add custom connector"** at the bottom of the connectors list

3. **Fill in the connector details:**
   - **Name**: `Design Systems Assistant` (or any name you prefer)
   - **URL**: `https://design-systems-mcp.southleft.com/mcp`

4. **Click "Add"** to save the connector

5. **Start using it!** The connector will appear in your connectors list with 4 available tools:
   - `search_design_knowledge`
   - `search_chunks`
   - `browse_by_category`
   - `get_all_tags`
   - `browse_by_tag`

**That's it!** You can now use the Design Systems Assistant in your Claude Desktop conversations.

**Note:** Custom connectors are available for Claude Pro, Team, and Enterprise plans.

Claude Code (CLI) - Click to expand configuration

**Quick Setup via CLI:**```bash
claude mcp add --transport http design-systems https://design-systems-mcp.southleft.com/mcp

Or manually edit .mcp.json:```json
{
"mcpServers": {
"design-systems": {
"type": "http",
"url": "https://design-systems-mcp.southleft.com/mcp"
}
}
}


**Verify connection:**```bash
claude mcp list

Cursor IDE - Click to expand configuration

Location: ~/.cursor/mcp_config.json or `~/.config/cursor/mcp_config.json````json
{
"mcpServers": {
"design-systems": {
"url": "https://design-systems-mcp.southleft.com/mcp"
}
}
}


**Restart Cursor** after updating the configuration.

Cline (VSCode Extension) - Click to expand configuration

**Location:** VSCode Settings → Extensions → Cline → MCP Settings

**Add to MCP servers configuration:**```json
{
  "design-systems": {
    "url": "https://design-systems-mcp.southleft.com/mcp",
    "description": "Design systems knowledge and best practices"
  }
}

Or add via Command Palette: Cline: Add MCP Server

Reload VSCode after configuration.

Continue (VSCode Extension) - Click to expand configuration

Location: VSCode Settings → Extensions → Continue → config.json```json
{
"mcpServers": [
{
"name": "design-systems",
"url": "https://design-systems-mcp.southleft.com/mcp",
"description": "Design systems knowledge base"
}
]
}


Zed Editor - Click to expand configuration

**Location:** `~/.config/zed/settings.json````json
{
  "mcp": {
    "servers": {
      "design-systems": {
        "url": "https://design-systems-mcp.southleft.com/mcp"
      }
    }
  }
}

Generic MCP Client - Click to expand configuration

For any MCP client supporting remote servers:

Endpoint: https://design-systems-mcp.southleft.com/mcp

Protocol: JSON-RPC 2.0 over HTTP/HTTPS

Transport: Standard MCP transport (stdio, SSE, or HTTP)

Local Development Setup - Click to expand configuration

To connect to your local development server instead of the public endpoint:```json
{
"mcpServers": {
"design-systems": {
"url": "http://localhost:8787/mcp"
}
}
}


**Note:** Local server requires running `npm run dev` first.

### Connection Troubleshooting

**Server not responding?**
- Verify the URL is correct: `https://design-systems-mcp.southleft.com/mcp`
- Test with curl: `curl https://design-systems-mcp.southleft.com/health`
- Check your client supports remote MCP servers

**Tools not appearing?**
- Restart your MCP client after configuration changes
- Check client logs for connection errors
- Verify JSON configuration syntax is correct

**Need help?**
- Open an issue: GitHub Issues

## Available MCP Tools

The server provides these tools for AI assistants:

### search_design_knowledge
Search the complete knowledge base with semantic understanding.

**Parameters:**
- `query` (string, required) - Search query
- `category` (string, optional) - Filter by category
- `tags` (array, optional) - Filter by tags
- `limit` (number, optional) - Max results (default: 15)

**Example:**```json
{
  "name": "search_design_knowledge",
  "arguments": {
    "query": "WCAG 2.2 color contrast requirements",
    "category": "guidelines",
    "limit": 5
  }
}

search_chunks

Find specific information within content chunks for detailed answers.

Parameters:
- query (string, required) - Search query
- limit (number, optional) - Max chunks (default: 8)

Example:```json
{
"name": "search_chunks",
"arguments": {
"query": "W3C DTCG design tokens specification",
"limit": 3
}
}


### browse_by_category
Browse content organized by category.

**Categories:** components, tokens, patterns, guidelines, workflows, general

**Parameters:**
- `category` (string, required) - Category to browse

### get_all_tags
Get all available content tags for filtering and exploration.

## API Examples

### Direct API Testing

**Health Check:**```bash
curl https://design-systems-mcp.southleft.com/health

MCP Tools List:```bash
curl -X POST https://design-systems-mcp.southleft.com/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'


**Search Query:**```bash
curl -X POST https://design-systems-mcp.southleft.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "search_chunks",
      "arguments": {"query": "design tokens", "limit": 3}
    }
  }'

AI Chat Interface (streaming):

The /ai-chat endpoint returns a Server-Sent Events stream so content appears progressively. Each event is data: {"t": ""}\n\n, terminated by event: done\ndata: {}\n\n.```bash
curl -N -X POST https://design-systems-mcp.southleft.com/ai-chat \
-H "Content-Type: application/json" \
-d '{"message":"What are the WCAG 2.2 contrast requirements?"}'


The hosted web UI at `/` consumes this stream and renders markdown progressively.

## Adding Content

### Ingest Web Content```bash
# Single URL
npm run ingest:url https://material.io/components/buttons

# Bulk from CSV
npm run ingest:csv urls.csv

# Crawl entire website
npm run crawl:website https://polaris.shopify.com --max-depth 3

Ingest PDF Content```bash

npm run ingest:pdf path/to/design-guide.pdf


### Generate Vector Embeddings```bash
npm run ingest:vectors

Development

Available Scripts

  • npm run dev - Start local development server
  • npm run deploy - Deploy to Cloudflare Workers
  • npm run ingest:pdf - Ingest PDF content
  • npm run ingest:url - Ingest web content
  • npm run ingest:csv - Bulk ingest from CSV
  • npm run crawl:website - Crawl entire websites
  • npm run ingest:vectors - Generate embeddings for all content
  • npm run setup:supabase - Initialize Supabase database
  • npm run check:duplicates - Check for duplicate content

Project Structure```

design-systems-mcp/
├── src/
│ ├── index.ts # Main MCP server, transports, tool dispatch, embedded chat UI
│ ├── sse-session.ts # SSE transport (Durable Object)
│ ├── streamable-http-handler.ts # Streamable HTTP transport (/mcp)
│ ├── oauth-handler.ts # OAuth flow
│ └── lib/
│ ├── content-manager.ts # Content management
│ ├── search-handler.ts # Vector + keyword search dispatch
│ ├── source-authority.ts # Reliability tiers & APG disclaimers
│ └── ... (chunker, formatters, ingestion helpers)
├── content/
│ └── entries/ # Ingested content (JSON)
├── supabase/
│ └── migrations/ # SQL schema + RPC functions
├── scripts/
│ ├── ingestion/ # Content ingestion pipeline (URL, PDF, HTML, CSV, crawler)
│ └── build/ # Build helpers (manifest generation)
├── types/
│ └── content.ts # TypeScript definitions
├── wrangler.jsonc # Cloudflare Workers config
└── .dev.vars # Local environment variables
```

Deployment

Deploy to Cloudflare Workers

  1. Login to Cloudflarebash npx wrangler login

  2. Set Secretsbash npx wrangler secret put OPENAI_API_KEY npx wrangler secret put SUPABASE_URL npx wrangler secret put SUPABASE_SERVICE_KEY npx wrangler secret put SUPABASE_ANON_KEY

  3. Deploybash npm run deploy

See DEPLOYMENT.md for detailed instructions.

Vector Search Architecture

This server uses Supabase for production-grade vector search:

  • Database: PostgreSQL with pgvector extension
  • Embeddings: Cloudflare Workers AI @cf/baai/bge-m3 (1024-dim, edge-native, no API key) via VECTOR_SEARCH_PROVIDER=cloudflare; OpenAI text-embedding-3-small (1536-dim) also supported
  • Threshold: 0.15 for optimal recall
  • Hybrid Search: Combines semantic vectors with text matching
  • Performance: Sub-100ms queries with proper indexing

Statistics:
- 395 entries in production database
- 4,800+ content chunks; entry-level vector search via Cloudflare Workers AI
- W3C standards, WCAG guidelines, design system documentation
- Regular updates with new authoritative sources

Troubleshooting

Common Issues

Vector search not working:
- Check Supabase credentials in environment variables
- Verify database tables exist: npm run setup:supabase
- Check logs: npx wrangler tail

Content not found:
- Verify content exists: npm run check:duplicates
- Check if embeddings generated: Look for embedding field in content entries
- Test search locally: npm run dev and use curl commands

MCP connection fails:
- Verify URL is correct and accessible
- Check client supports remote MCP servers
- Test with curl: curl https://design-systems-mcp.southleft.com/health
- Restart MCP client after configuration changes

Documentation

  • DEPLOYMENT.md - Production deployment
  • CONTRIBUTING.md - How to contribute and add content
  • CREDITS.md - Content sources and attribution

License & Attribution

License: MIT License - Free for personal and commercial use

Content Attribution: This project compiles design systems knowledge from many brilliant creators. All original content remains the intellectual property of their respective authors.

  • See CREDITS.md for complete attribution
  • Always link back to original sources when sharing insights
  • Support original creators by visiting their websites

Security & Privacy

  • No sensitive data stored - Only public design system knowledge
  • Environment variables use Cloudflare secrets
  • Open source and auditable
  • Privacy-focused - No user data collection
  • Regular security updates

Report security issues to: GitHub Security

Contributing

We welcome contributions! Whether you want to:
- Report bugs or issues
- Suggest new features
- Add more design system content
- Improve the codebase
- Enhance documentation

Please:
1. Check existing issues
2. Open a new issue to discuss
3. Submit a pull request
4. Follow contribution guidelines

Support

  • Issues: GitHub Issues
  • Discussions: GitHub Discussions
  • Live Demo: https://design-systems-mcp.southleft.com/

Acknowledgments

Thanks to the design systems community for sharing knowledge:

  • Brad Frost for Atomic Design methodology
  • W3C Design Tokens Community Group
  • Web Accessibility Initiative (WAI)
  • All design teams who openly share their work
  • The entire design systems community

See CREDITS.md for the complete list.


Built with ❤️ using Cloudflare Workers and the Model Context Protocol

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
Design Systems Mcp 怎么安装?

在客户端按下面的安装命令接入:https://design-systems-mcp.southleft.com/mcp。装完在客户端配置里引用同一个包或端点。

Design Systems Mcp 怎么在 Claude Desktop、Claude、Cursor 里配置?

它的传输方式是HTTP · 远程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Design Systems Mcp 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Design Systems Mcp 需要 API Key 吗?

资料标注需要OAuth 授权认证,使用前准备好对应凭据,并按最小权限配置。

Design Systems Mcp 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、VS Code、Windsurf、Cline;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Design Systems Mcp 能提供哪些能力?

资料显示能力标签包括 mcp、ai assistant。具体工具清单以官方文档为准。

Design Systems Mcp 需要什么运行环境?

运行时为 node;属于远程端点,不需要在本地装运行时。

Design Systems Mcp 是本地还是远程?

它是远程端点,连服务地址即可。传输方式为HTTP · 远程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 未标注。