Integrate an AI agent with Foundry IQ

Why it matters

Foundry IQ is the managed path between an agent and enterprise knowledge, and this exercise shows what that path exposes to the application that calls the agent.

What to practise

Create an agent in the Foundry portal and connect it to Foundry IQ, building a knowledge base over product PDFs in Blob Storage, backed by an Azure AI Search resource. In the Foundry Toolkit for VS Code, set the kb-knowledgebase… tool to never auto-approve. Then complete a Python client that uses the conversations API and loops over mcp_approval_request items until none remain. Watch for citations, for a follow-up that relies on earlier turns, and for the answer when the knowledge base has nothing. The exercise notes that some of its technologies are in preview.

How it fits

The knowledge base reached as an MCP tool is 2.2.12’s integration choice, compared against direct Search. The service-side conversation is 2.2.1’s single owner of continuity; the client’s local history list is only for display. The approval loop previews 2.3.4.