Google Shopping Agent Demo

[yenayuu]

Feature: Shopping Agent MCP Tool

Feature Description

Finding and comparing products across multiple retailers is time-consuming and overwhelming. Users need personalized recommendations that consider their budget, preferences, and requirements without manually browsing dozens of websites.

Proposed Solution

Implement a Shopping Agent MCP server that provides AI-powered product recommendations by orchestrating multi-step workflows using LangGraph, automatically parsing queries, searching across retailers, and generating curated suggestions.

Workflow

The agent will do the following:

• Accept natural language shopping queries (e.g., "I want to buy a scarf for 40 dollars")
• Parse query to extract product type and budget constraints
• Search across multiple retailers using real-time product search
• Analyze results and generate personalized recommendations with reasoning
• Return structured JSON with product details, prices, and purchase links

Specifications

MCP Tools

recommend_products

• Input: query (string), maxResults (integer, default: 5, max: 20)
• Output: JSON array with product name, price, description, URL, and recommendation reason
• Uses LangGraph multi-node workflow for intelligent processing

search_products

• Input: query (string), maxResults (integer, default: 10, max: 100)
• Output: Raw search results for advanced users
• Lower-level tool for custom processing

LangGraph Workflow

parse_query → search_products → generate_recommendations → END

Three-node state graph with typed state management:

• messages: Conversation history with message reducer
• search_results: Raw product data from searches
• recommendations: Final curated product list

Configuration

# Required environment variables
SERPAPI_API_KEY="your-key"

# Optional configuration
HOST="0.0.0.0"              # Server host
PORT="8000"                  # Server port
MCP_TRANSPORT="http"         # Transport type
LOG_LEVEL="INFO"             # Logging level

Example Query/Response

Input:

{
  "query": "wireless headphones under $100 with noise cancellation",
  "maxResults": 5
}

Output:

{
  "query": "wireless headphones under $100 with noise cancellation",
  "recommendations": [
    {
      "name": "Sony WH-CH710N",
      "price": "$88.00",
      "description": "Wireless noise-canceling headphones with 35hr battery",
      "url": "https://...",
      "reason": "Excellent noise cancellation within budget, high ratings"
    }
  ],
  "count": 5
}

Testing Modes

• MCP Inspector: Connect to http://localhost:8000 for visual testing
• Direct Testing: Run simple_test.py to test workflow without MCP layer
• Client Testing: Use test_client.py for protocol-level testing

Deployment

• Docker support with single-command deployment
• Environment-based configuration
• Stateless design for horizontal scaling

APIs

• SerpAPI: Real-time product search across multiple retailers
• LangChain: LLM framework for agent orchestration and message management
• LangGraph: Multi-step workflow with state management and node execution
• SentenceTransformers: (Future) Enhanced product categorization and similarity matching
• FastMCP: MCP server framework for tool exposure and protocol handling

Documentation

• README.md: Complete setup and usage guide
• ARCHITECTURE.md: System design with data flow diagrams (449 lines)
• QUICKSTART.md: Step-by-step getting started guide
• TROUBLESHOOTING.md: Common issues and MCP protocol explanations

[mrsabath]

Closing as it has been addressed by PR #100