Foundry Agents Samples

End-to-end Python samples for building and running AI Agents with Azure AI Foundry Agent Service using the azure-ai-projects v2 SDK (>=2.0.0).

Prerequisites

• Python 3.10+
• An Azure AI Foundry project with at least one LLM deployment
• The Azure CLI installed and signed in (az login) — used for DefaultAzureCredential

Samples

#	Folder	Tool	Description
01	01-search-tool-agent	AzureAISearchTool	Agent that queries an Azure AI Search index directly for grounded, citation-backed answers. Simplest search integration — no knowledge base required.
02	02-mcp-tools/foundry-iq	MCPTool (Foundry IQ)	Agent that connects to a Foundry IQ knowledge base via MCP for agentic retrieval: LLM-based query planning, parallel subqueries, semantic reranking, and answer synthesis.
03	03-code-interpreter-agent	Code Interpreter	Agent that uses the built-in code interpreter tool to analyze data, run Python, and produce results programmatically.
04	04-model-gtw	ModelGateway	Agent backed by a ModelGateway connection to route inference through an external AI gateway (e.g. APIM AI Gateway) using OAuth2 client credentials.
05	05-openapi-tool/apim	OpenApiTool	Agent that calls an Azure API Management endpoint described by an OpenAPI spec. APIM subscription key is stored as a Foundry CustomKeys connection and injected automatically at runtime — no credentials in code.

Search Tool vs Foundry IQ: Both use Azure AI Search but differ in retrieval depth. AzureAISearchTool queries the index directly — simpler, faster to set up. Foundry IQ adds an agentic pipeline on top (query decomposition, reranking, synthesis) — better for complex queries and multi-source scenarios. See 01-search-tool-agent/README.md for a full comparison table.

Structure

Each sample is self-contained under its own numbered folder:

<sample-folder>/
├── config.json.example  # Template — copy to config.json and fill in your values
├── config.json          # Your local configuration (git-ignored, never committed)
├── requirements.txt     # Python dependencies for this sample
├── README.md            # Setup and run instructions specific to this sample
└── *.py                 # Sample script(s)

Authentication

All samples use DefaultAzureCredential, which supports:

• Local development: az login (Azure CLI)
• Production / hosted: Managed Identity

No API keys are used or required.

Validation environment

These samples have been validated end-to-end in a private Azure AI Foundry environment with:

• VNet-injected agents (no public network access on the Foundry resource)
• Private endpoints for Azure AI Search and storage
• All traffic routed over the private network

If you are running in a public (non-VNet) environment the samples work the same way — the VNet setup only affects network routing, not the SDK code or configuration structure.