Design Systems MCP Server

An AI-powered Model Context Protocol (MCP) server that provides intelligent access to design systems knowledge. This server ingests design system documentation (PDFs, web content) and enables AI assistants to provide expert guidance on design systems, components, tokens, and best practices.

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

Features

🤖 AI-Powered Chat Interface - Natural language queries with OpenAI integration
📚 Content Ingestion - Supports PDF parsing and web content extraction
🔍 Intelligent Search - Semantic search across design systems documentation
🎨 Rich Formatting - Markdown rendering with syntax highlighting
🚀 Cloudflare Workers - Scalable serverless deployment
🧪 Local Testing - Full local development environment
🌐 Public Access - Live MCP server available for external integrations

Live MCP Server

🚀 Public Endpoints

Workers Domain:https://design-systems-mcp.southleft-llc.workers.dev

AI Chat Interface:https://design-systems-mcp.southleft.com/
MCP Endpoint:https://design-systems-mcp.southleft-llc.workers.dev/mcp
Health Check:https://design-systems-mcp.southleft-llc.workers.dev/health

✨ Try It Now

Visit the live demo and ask questions like:

"What are design tokens and how should I use them?"
"How do I create accessible button components?"
"What are the best practices for organizing a design system?"
"How do components work in design systems?"

Quick Start

Prerequisites

Node.js (v20.17.0+ or v22.9.0+)
OpenAI API key (for local development)

Local Development Setup

Clone and Install

git clone https://github.com/southleft/design-systems-mcp.git
cd design-systems-mcp
npm install

Configure Environment

cp env.example .dev.vars
# Edit .dev.vars and add your OpenAI API key

Start Development Server
```
npm run dev
```
Server will be available at: http://localhost:8787
Test the AI Chat Interface
- Open http://localhost:8787 in your browser
- Try example queries like:
  - "What are design tokens?"
  - "Where in the design systems handbook is Alicia SEDLOCK mentioned?"
  - "What does the Design Systems Handbook say about Wraith, Gemini, and BackstopJS?"
  - "How do I implement a design system?"

Adding Content

Ingest Content (if not already done)

# Add PDFs to local-content-library/
npm run ingest:pdf path/to/your-design-guide.pdf

# Or ingest web content
npm run ingest:url https://example.com/design-system

# Or bulk ingest from CSV file
npm run ingest:csv path/to/urls.csv

Update Content Loading in src/index.ts

// Add new content files using dynamic imports
const [handbookModule, buttonModule, newContentModule] = await Promise.all([
  import('../content/entries/8zWJWrDK_bTOv3_KFo30V-pdf-designsystemshandbook-pdf.json'),
  import('../content/entries/sample-button-guidelines.json'),
  import('../content/entries/your-new-content.json')
]);

const actualEntries = [
  handbookModule.default as ContentEntry,
  buttonModule.default as ContentEntry,
  newContentModule.default as ContentEntry
];

Test Locally

npm run dev
# Test your new content in the chat interface

Available Tools

The MCP server provides these tools for AI assistants:

search_design_knowledge - Search design systems content
search_chunks - Find specific information in content chunks
browse_by_category - Browse content by category (components, tokens, etc.)
get_all_tags - Get available content tags

Local Testing Workflow

Testing New Content

Add content files to content/entries/
Update src/index.ts to load new content
Restart dev server: npm run dev
Test queries in chat interface at http://localhost:8787
Verify AI responses are accurate and complete

Testing MCP Tools Directly

Local Testing:

# Test MCP search directly
curl -X POST http://localhost:8787/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_chunks","arguments":{"query":"design tokens"}}}'

# Test AI integration
curl -X POST http://localhost:8787/ai-chat \
  -H "Content-Type: application/json" \
  -d '{"message":"What are design tokens?"}'

Production Testing:

# Test live MCP endpoint
curl -X POST https://design-systems-mcp.southleft-llc.workers.dev/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_chunks","arguments":{"query":"design tokens"}}}'

# Test live AI integration
curl -X POST https://design-systems-mcp.southleft-llc.workers.dev/ai-chat \
  -H "Content-Type: application/json" \
  -d '{"message":"What are design tokens?"}'

Deployment

Deploy to Cloudflare Workers

Login to Cloudflare
```
npx wrangler login
```

Set Environment Variables

npx wrangler secret put OPENAI_API_KEY
# Enter your OpenAI API key when prompted

# Optional: Set custom model (current default: gpt-4o)
npx wrangler secret put OPENAI_MODEL
# Enter model name (default: gpt-4o)

Deploy
```
npm run deploy
```
Access Your Deployed Server
- Your server will be available at: design-systems-mcp.<your-account>.workers.dev
- Chat interface: https://design-systems-mcp.<your-account>.workers.dev
- MCP endpoint: https://design-systems-mcp.<your-account>.workers.dev/mcp

Custom Domain Setup

To set up a custom domain like design-systems-mcp.southleft.com:

Deploy your worker (see steps above)
In Cloudflare Dashboard:
- Go to Workers & Pages → Custom Domains
- Add your custom domain
- Point it to your deployed worker: design-systems-mcp
Configure DNS in your domain settings
Test the endpoints once propagated

Environment Variables

Required environment variables:

OPENAI_API_KEY - Your OpenAI API key
OPENAI_MODEL - Model to use (default: "gpt-4o")
AI_SYSTEM_PROMPT - Custom system prompt (optional)

Connect to MCP Clients

Claude Desktop / Cursor

Add to your MCP configuration file (~/.cursor/mcp.json for Cursor, or claude_desktop_config.json for Claude):

Option 1: Use Public Remote Server (Recommended for most users)

{
  "mcpServers": {
    "design-systems": {
      "url": "https://design-systems-mcp.southleft-llc.workers.dev/mcp"
    }
  }
}

Option 2: Use Local Development Server (For contributors/customization)

{
  "mcpServers": {
    "design-systems": {
      "url": "http://localhost:8787/mcp"
    }
  }
}

Important Notes:

✅ Both local and remote servers are fully functional
🌐 Remote server: Always available, no setup required
🔧 Local server: Requires running npm run dev first
📝 After updating configuration, restart your MCP client (Cursor/Claude)

Cloudflare AI Playground

Go to https://playground.ai.cloudflare.com/
Enter your MCP server URL: design-systems-mcp.southleft-llc.workers.dev/mcp
Start using design systems tools in the playground!

External Applications

Any application that supports MCP can connect to the live server:

Endpoint:https://design-systems-mcp.southleft-llc.workers.dev/mcp

Example API Call:

# Initialize connection
curl -X POST https://design-systems-mcp.southleft-llc.workers.dev/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {"roots": {"listChanged": true}},
      "clientInfo": {"name": "test", "version": "1.0.0"}
    }
  }'

# Search design systems knowledge
curl -X POST https://design-systems-mcp.southleft-llc.workers.dev/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "search_design_knowledge",
      "arguments": {"query": "design tokens"}
    }
  }'

Project Structure

design-systems-mcp/
├── src/
│   ├── index.ts              # Main server with AI integration
│   ├── lib/
│   │   └── content-manager.ts # Content management and search
│   └── tools/                # MCP tool definitions
├── content/
│   ├── entries/              # Ingested content (JSON)
│   └── raw/                  # Raw source files
├── scripts/
│   └── ingestion/            # Content ingestion scripts
├── types/
│   └── content.ts           # TypeScript definitions
├── local-content-library/   # Source PDFs and files
├── wrangler.jsonc          # Cloudflare Workers config
└── .dev.vars              # Local environment variables

Content Management

Supported Content Types

PDFs - Design system handbooks, guidelines
Web Content - Design system documentation sites
CSV URLs - Bulk ingestion from CSV files containing multiple URLs
JSON - Pre-processed design system data

CSV Bulk Ingestion

For bulk content ingestion, you can use CSV files containing multiple URLs:

1. Create a CSV File

# Generate a sample CSV template
npm run ingest:csv --sample

2. CSV Format

Your CSV file should include these columns (header row recommended):

Column	Required	Description
`url`	✅	The URL to fetch content from
`title`	⚪	Custom title for the content
`category`	⚪	Content category (general, components, tokens, patterns, guidelines, tools)
`tags`	⚪	Comma-separated tags
`description`	⚪	Description of the content
`confidence`	⚪	Confidence level (low, medium, high)
`system`	⚪	Design system name
`author`	⚪	Author or organization
`version`	⚪	Version information

3. Example CSV

url,title,category,tags,description,confidence,system,author,version
https://material.io/components/buttons,Material Design Buttons,components,"button,interaction,material",Material Design button guidelines,high,Material Design,Google,3.0
https://polaris.shopify.com/components/button,Shopify Polaris Button,components,"button,shopify,polaris",Shopify's button component,high,Polaris,Shopify,
https://primer.style/components/button,GitHub Primer Button,components,"button,github,primer",GitHub's button guidelines,high,Primer,GitHub,

4. Ingest Content

# Basic ingestion
npm run ingest:csv my-urls.csv

# With custom options
npm run ingest:csv my-urls.csv --max-concurrent 5 --timeout 60000

# Dry run (validate without fetching)
npm run ingest:csv my-urls.csv --dry-run

# See all options
npm run ingest:csv --help

5. Advanced Options

--max-concurrent <n> - Process N URLs simultaneously (default: 3)
--timeout <ms> - Request timeout in milliseconds (default: 30000)
--retry-attempts <n> - Number of retry attempts for failed URLs (default: 2)
--output-dir <dir> - Custom output directory (default: content/entries)
--delimiter <char> - CSV delimiter (default: ',')
--no-header - CSV file doesn't have a header row

Content Processing

Content is automatically:

Chunked for optimal search performance
Tagged and categorized
Indexed for semantic search
Made available to AI for intelligent responses

Development

Available Scripts

npm run dev - Start local development server
npm run deploy - Deploy to Cloudflare Workers
npm run ingest:pdf <file> - Ingest PDF content
npm run ingest:url <url> - Ingest web content
npm run ingest:csv <file> - Bulk ingest from CSV file containing URLs
npm run check:duplicates - Check for duplicate URLs in content entries

Content Quality Assurance

Check for Duplicate URLs:

npm run check:duplicates

This command scans all content entries and identifies any duplicate URLs to maintain content quality. Run this:

Before deploying new content
After ingesting new articles
Periodically to ensure data integrity

The checker will show:

Total entries scanned
Number of unique URLs found
Any duplicates with filenames and titles
Suggested cleanup commands

Adding New MCP Tools

Define tools in src/index.ts:

server.tool("your_tool_name", schema, async (params) => {
  // Tool implementation
});

Add to OpenAI function definitions:

const MCP_TOOLS = [
  // ... existing tools
  {
    type: "function",
    function: {
      name: "your_tool_name",
      description: "Tool description",
      parameters: { /* JSON schema */ }
    }
  }
];

Troubleshooting

Common Issues

Content not loading:

Check that JSON files exist in content/entries/
Verify dynamic import paths in src/index.ts
Check server logs for loading errors
Ensure content files are valid JSON format

Port issues:

Ensure wrangler.jsonc has correct dev port (8787)
Kill existing processes: pkill -f "wrangler dev"

Environment variables:

Local: Use .dev.vars file
Production: Set via npx wrangler secret put

Logs and Debugging

# View server logs
npx wrangler tail

# Local development logs
npm run dev
# Check console output for content loading status

📄 License & Usage

This project is free and open source under the MIT License. You are welcome to:

✅ Use it for personal and commercial projects
✅ Modify and distribute it
✅ Build upon it for your own projects
✅ Share it with your team and community

Content Attribution

This project compiles design system 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 the original creators by visiting their websites and platforms

🔒 Security & Privacy

No sensitive data is stored - Only public design system knowledge
Environment variables are secure - API keys use Cloudflare secrets
Open source and auditable - All code is publicly available
Privacy-focused - No user data collection beyond basic usage analytics

See SECURITY.md for detailed security information and best practices.

🤝 Contributing

We welcome contributions! Whether you want to:

🐛 Report bugs or issues
💡 Suggest new features or improvements
📚 Add more design system content
🔧 Improve the codebase
📖 Enhance documentation

Please:

Check existing issues first
Open a new issue to discuss your idea
Submit a pull request with your changes
Follow our security guidelines

Adding Content

To contribute new design system content:

Ensure you have permission to share the content
Follow the ingestion process documented above
Add proper attribution in CREDITS.md
Submit a pull request with the new content

🙏 Acknowledgments

This project exists thanks to the generous sharing of knowledge from the design systems community. Special thanks to:

Brad Frost for the foundational Atomic Design methodology
The Design System Guide team for comprehensive practical resources
Figma for excellent official documentation
All the design teams who openly share their experiences and methodologies
The entire design systems community for fostering knowledge sharing

See CREDITS.md for the complete list of contributors and sources.

📞 Support & Community

🐛 Issues:GitHub Issues
📧 Security: Report security issues privately to the maintainers
🌐 Website:Live Demo

Legacy Cloudflare Template Information

This project was built from the Cloudflare remote MCP server template. For additional Cloudflare Workers information:

Original Template Deploy

Command Line Template

npm create cloudflare@latest -- my-mcp-server --template=cloudflare/ai/demos/remote-mcp-authless

Server	Summary	Actions
Hostinger Ecommerce MCP	This is a Model Context Protocol (MCP) server for interacting with the Hostinger Ecommerce API.	View
Google Ads		View
AWS Customer Playbook Advisor MCP	※ このリポジトリは非公式のものであり、クローンしてもらって試してもらっても良いですが責任は負いません...	View
Salesforce Einstein by CData	CData's Model Context Protocol (MCP) Server for Salesforce Einstein	View
Axiom	A Model Context Protocol server implementation for Axiom that enables AI agents to query your data u...	View
AWS MCP	A Model Context Protocol (MCP) server that enables AI assistants like Claude to interact with your A...	View

Cloudflare MCP Server Template

Design Systems MCP Server

Features

Live MCP Server

🚀 Public Endpoints

✨ Try It Now

Quick Start

Prerequisites

Local Development Setup

Adding Content

Available Tools

Local Testing Workflow

Testing New Content

Testing MCP Tools Directly

Deployment

Deploy to Cloudflare Workers

Custom Domain Setup

Environment Variables

Connect to MCP Clients

Claude Desktop / Cursor

Cloudflare AI Playground

External Applications

Project Structure

Content Management

Supported Content Types

CSV Bulk Ingestion

1. Create a CSV File

2. CSV Format

3. Example CSV

4. Ingest Content

5. Advanced Options

Content Processing

Development

Available Scripts

Content Quality Assurance

Adding New MCP Tools

Troubleshooting

Common Issues

Logs and Debugging

📄 License & Usage

Content Attribution

🔒 Security & Privacy

🤝 Contributing

Adding Content

🙏 Acknowledgments

📞 Support & Community

Legacy Cloudflare Template Information

Original Template Deploy

Command Line Template

Related in Cloud Service - Secure MCP Servers