Mission Pizza: AI-Ready MCP Architecture

🚀 The Solution

This project automates the transition from legacy OpenAPI specs to the Model Context Protocol (MCP). Instead of manually writing tools, I built a system that generates them dynamically.

🏗 Architecture

• Legacy System: A Mock FastAPI backend (simulating the pizza database).
• The Bridge (generate_mcp.py): A Python script that parses openapi.json and auto-generates a fully compliant MCP server (server.py). This ensures scalability—if the menu changes, the AI tools update automatically.
• Agent Swarm:
  • Ordering Agent: Uses Gemini 1.5 to interact with the generated MCP server.
  • Scheduling Agent: Receives order context (ETA) and uses a simulated Calendar MCP to block time for a "Pizza Study Break.

🛠 How to Run

1. Install Dependencies:

   pip install fastapi uvicorn mcp google-genai python-dotenv

2. Start the Backend:

   uvicorn backend:app --port 8000

3. Generate the MCP Server:

   python backend.py generate_spec   # Creates json
   python generate_mcp.py            # Creates server.py

4. Run the Multi-Agent Workflow:

   python main_flow.py

🧠 Design Decisions

• Why Mock Data? As per the prompt requirements, I focused on the MCP transformation logic rather than building a complex SQL backend.
• Security: API Keys are managed via .env (not committed).
• Robustness: The generator sanitizes function names to prevent Python syntax errors during the JSON-to-MCP conversion.

System Data Flow

graph TD
    User[User] -->|Natural Language| Agent1[Ordering Agent]
    Agent1 -->|Tools| MCPServer[Generated MCP Server]
    MCPServer -->|API Calls| Backend[Legacy FastAPI Backend]
    
    Agent1 -->|Order Context| Agent2[Scheduling Agent]
    Agent2 -->|Tools| Calendar[Calendar MCP]
    
    subgraph "The Innovation Layer"
        OpenAPI[OpenAPI Spec] -->|Parsed by| Generator[generate_mcp.py]
        Generator -->|Writes Code| MCPServer
    end

Loading