How microsoft/ignite25-lab571-build-a-pizza-ordering-agent-with-microsoft-foundry-and-mcp works
Project type and purpose
• This is not a standard web or mobile app, but a hands-on lab/tutorial repository for “LAB571: Build a Pizza Ordering Agent with Microsoft Foundry and MCP.”
• The goal is to demonstrate how to build, prompt-tune, and extend a domain-specific AI agent (the Contoso PizzaBot) using:
• Microsoft Foundry Agent Service (for hosting and running the agent)
• Python (the implementation language via the Foundry SDK)
• RAG (Retrieval-Augmented Generation) for knowledge
• Tool calling (e.g., pizza calculator)
• MCP (Microsoft Cloud Platform) for live menu and order handling
File structure
• Top-level config:
• .devcontainer/, .vscode/ – development environment settings (VS Code + Docker)
• .github/ – CI/workflow and contributor guidance
• requirements.txt – Python dependencies (Foundry SDK, Azure SDKs, etc.)
• Documentation & lab flows:
• lab/1.md–lab/6.md – step-by-step workshop modules
• lab/foundry/ – likely the sample code that students will run
• AGENTS.md – details on agent definitions
• No “frontend” or “backend” folders per se, because this is a service-based AI agent with no UI code in this repo.
README for additional insights
• Confirms the session teaches: system prompts, custom instructions, RAG, tool calling, MCP integration
• Mentions Python and the Foundry SDK as the technical stack
Main components
a. Developer Environment
• VS Code (local machine / devcontainer)
• Python + Foundry SDK (requirements.txt)
• GitHub actions (for CI, automated lab validation)
b. Agent Service (Hosted in Microsoft Foundry)
• Core “PizzaBot” agent definition (system prompt, user prompt flows)
• Components: Prompt templates, RAG retriever, tool‐calling interface
c. Tools & Skills
• Pizza Calculator (Python function or microservice)
• Additional skill modules as defined in AGENTS.md
d. Knowledge Store (RAG)
• Vector store / embeddings (e.g., Azure Cognitive Search + Azure OpenAI)
• Source documents (menu, FAQs) loaded at runtime
e. External Services (via MCP)
• Live menu API (Contoso Pizza Menu Service)
• Order Management API (place, update, retrieve orders)
f. End Users
• Chat interface (CLI, custom frontend, or VS Code extension) interacting with the running agent
Relationships and interactions
• Developer pushes code (labs & agent definitions) → CI/CD → Foundry Agent Service
• End user sends a query → Foundry Agent Service
→ system prompt + user prompt processed
→ RAG retriever queries vector store for domain knowledge
→ tool‐calling layer invokes Pizza Calculator or MCP APIs
→ Aggregated response returned to user
• Any data or telemetry flows back to Azure (monitoring/logging)
Architectural patterns & design principles
• Modular “Skill” or “Tool” pattern for each unit of functionality (calculator, order API)
• RAG pattern: separation of prompt logic and knowledge retrieval
• Service-oriented: the agent orchestrates external microservices rather than hard-coding logic