Nike Search MCP Server

A Model Context Protocol (MCP) server that provides real-time Nike product search and comparison capabilities with custom widgets. Built following the official OpenAI Pizzaz Python example patterns using FastMCP.

Works with ChatGPT (via Connectors), Claude Desktop, and any MCP-compatible client.

✨ Features

• 🔍 Real Nike Product Data - Scrapes Nike.com for live product information
• 🖼️ Product Images - High-resolution images from Nike's CDN
• 💰 Pricing & Availability - Real-time pricing and stock information
• ⚖️ Product Comparison - Side-by-side product comparisons
• 🎨 Custom UI Widget - React component for rich product displays
• 🌐 Multiple Transports - Supports stdio (desktop) and SSE (web)

🚀 Quick Start

1. Install Dependencies

# Clone or navigate to the project
cd nikesearch

# Install Python dependencies (using uv)
uv sync

# Or use pip
pip install -r requirements.txt

2. Start the Server

# Run the FastMCP server
uv run python main.py

# Or directly with Python
python main.py

The server will start on http://127.0.0.1:8000 with the MCP endpoint at /mcp.

3. Connect to ChatGPT

Option A: Using ngrok (for ChatGPT Web Connectors)

1. Start ngrok in a new terminal:

   ngrok http 8000

2. Copy the ngrok HTTPS URL (e.g., https://abc123.ngrok-free.app)

3. In ChatGPT:

   • Go to Settings → Connectors
   • Click "Create new connector"
   • Enter your ngrok URL + /mcp:

     https://abc123.ngrok-free.app/mcp
     

   • Select "No Authentication"
   • Wait for validation

4. Test with: "Search for Nike Vomero"

See QUICK_START.md for more options.

Option B: Claude Desktop ✅ PROVEN

Works perfectly with stdio transport:

1. Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

   {
     "mcpServers": {
       "nike-search": {
         "command": "uv",
         "args": [
           "--directory",
           "/Users/ntorwe/nikesearch",
           "run",
           "nikesearch.py"
         ]
       }
     }
   }

2. Restart Claude Desktop

3. Ask: "Compare Nike Vomero and Pegasus"

Option C: ChatGPT Web Connectors (Experimental)

1. Start servers (REST API + ngrok)
2. Go to: https://chatgpt.com/#settings/Connectors
3. Add your ngrok URL

Note: May timeout; use Option A if issues occur.

📁 Project Structure

nikesearch/
├── main.py                    # ⭐ FastMCP server (Pizzaz-style)
├── requirements.txt           # Python dependencies
├── pyproject.toml             # UV project config
├── uv.lock                    # Locked dependencies
├── web/                       # React widget source (optional)
│   ├── src/
│   │   ├── component.tsx      # Main comparison widget
│   │   ├── ProductCard.tsx    # Product display component
│   │   └── types.ts           # TypeScript types
│   ├── dist/
│   │   └── component.js       # Bundled widget
│   └── package.json           # Node dependencies
└── docs/
    ├── README.md                       # This file
    ├── QUICK_START.md                  # Fast reference
    ├── PIZZAZ_REFACTOR_COMPLETE.md     # Refactor guide
    ├── READY_FOR_CHATGPT.md            # Setup guide
    └── CHATGPT_DEVELOPER_MODE_SETUP.md # Detailed instructions

Legacy files (kept for reference):
├── nikesearch.py              # Original MCP implementation
├── nikesearch_rest.py         # Old REST wrapper
└── test_mcp_client.py         # Test suite

🛠️ Available Tools

The MCP server provides these tools:

ChatGPT Connectors Tools (Required by OpenAI)

1. search - Search for Nike products

   • Input: query (string)
   • Returns: JSON with array of {id, title, url}

2. fetch - Fetch specific product details

   • Input: url (string - product URL or ID)
   • Returns: JSON with {id, title, text, url, metadata}

Full MCP Tools (For Claude & Apps SDK)

3. search_products - Advanced search with filters

   • Inputs: query, limit, gender, category
   • Returns: Detailed product list

4. get_product_details - Get full product information

   • Input: product_id
   • Returns: Complete product details

5. get_product_images - Get product image URLs

   • Input: product_id
   • Returns: Array of image URLs

6. compare_products - Side-by-side comparison

   • Inputs: product1, product2
   • Returns: Formatted comparison (with custom UI support)

🧪 Testing

Run the test suite:

uv run python test_mcp_client.py

This will verify:

• REST API endpoints
• MCP SSE connection
• All required manifests
• Tool availability

📖 API Documentation

Once the REST API server is running, visit:

• Interactive Docs: http://localhost:8000/docs
• OpenAPI Spec: http://localhost:8000/openapi.json
• Health Check: http://localhost:8000/health

🌐 Endpoints

Endpoint	Purpose
/	API information
/health	Health check
/api/search	REST search endpoint
/api/compare	REST compare endpoint
/.well-known/ai-plugin.json	AI plugin manifest
/app-manifest.json	ChatGPT Apps manifest
/openapi.json	OpenAPI specification
/sse	MCP Server-Sent Events
/docs	Interactive API docs

🎯 Example Queries

Once connected to ChatGPT or Claude:

• "Search for Nike Vomero"
• "Compare Nike Vomero and Pegasus"
• "Show me Nike running shoes"
• "What's the price of Nike Air Max?"
• "Find Nike basketball shoes"

🔧 Development

Build the React Widget

cd web
npm install
npm run build

This generates web/dist/component.js which is embedded in the MCP server.

Run Servers Locally

# Terminal 1: REST API Server
uv run python nikesearch_rest.py

# Terminal 2: ngrok (for HTTPS)
ngrok http 8000

# Terminal 3: Test
uv run python test_mcp_client.py

📚 Documentation

• QUICK_START.md - Fast reference for getting started
• READY_FOR_CHATGPT.md - Complete setup guide
• CHATGPT_DEVELOPER_MODE_SETUP.md - Detailed instructions

🔐 Authentication

Currently uses no authentication (auth: none) for simplicity. For production use, consider implementing:

• OAuth 2.0 dynamic client registration
• API key authentication
• Rate limiting

⚠️ Legal & Terms of Service

This tool scrapes Nike.com for educational and personal use. Please:

• Respect Nike's Terms of Service
• Implement rate limiting
• Cache results to minimize requests
• Do not use for commercial purposes without permission

🛡️ Privacy

• No user data is collected or stored
• All searches are performed in real-time
• No analytics or tracking
• See privacy_policy_url in manifests for details

🤝 Contributing

This is a personal project. Feel free to fork and customize for your own use.

📝 License

MIT License - See project for details.

🙏 Acknowledgments

• Built with Model Context Protocol
• Uses FastAPI for REST API
• React widget built with TypeScript
• Product data from Nike.com

---

Status: ✅ Ready for use with ChatGPT and Claude Desktop

Last Updated: October 13, 2025