Yina Arenas details how to build and securely manage agentic AI solutions in Azure AI Foundry, offering practical patterns, protocols, and governance strategies for enterprise adoption.

Agent Factory: Building Your First AI Agent with Azure AI Foundry

Author: Yina Arenas

Introduction

Agents’ capabilities are determined not only by the tools available to them, but by the governance and extensibility frameworks behind those tools. This article from the ‘Agent Factory’ series outlines how to build real-world agentic AI solutions in Microsoft Azure, using open protocols, reusable toolchains, and enterprise-grade management.

Defining the Next Wave of Agentic AI

The evolution of agentic AI is moving from simple, static prompts to broader extensibility. Developers are now expected to integrate agents with a wide range of APIs, services, and workflows, while maintaining strong governance and security.

Key challenges with early agent development included duplicated effort, brittle integrations, and fragmented governance. The adoption of open standards, such as the Model Context Protocol (MCP), addresses these challenges by providing self-describing, discoverable tools that agents can use at runtime across environments.

Azure AI Foundry supports MCP, alongside robust security, observability, and management, making it easier to build portable and compliant agentic solutions.

Building an Enterprise Toolchain in Azure AI Foundry

Azure AI Foundry provides three layers of extensibility for enterprise agentic AI:

  1. Built-in Tools: Pre-integrated tools for rapid value, such as searching SharePoint, running Python scripts, or automating browser tasks.
  2. Custom Tools: Integrate proprietary APIs or external systems using OpenAPI or MCP, with policy and observability layers built in.
  3. Connectors: Use Azure Logic Apps to connect agents to over 1,400 SaaS and on-premises systems.

Case Study: NTT DATA used Azure AI Foundry to integrate Microsoft Fabric Data Agent and reduce time-to-market by 50%.

Enterprise-Grade Management and Security

Azure AI Foundry takes a secure-by-default approach:

  • Authentication and Identity: Built-in connectors use Microsoft Entra ID (formerly Azure AD) for managed identity, ensuring authorization and compliance.
    • Microsoft Entra Agent ID centralizes management of AI agents across Azure environments.
    • Supports federation with external identity providers.
  • Governance with Azure API Management (APIM): Centralizes publishing, enforces policy, monitors usage, and enables self-hosted gateways for hybrid environments.
    • Azure API Center acts as an inventory and discovery hub for APIs and MCP servers.
  • Observability: All tool invocations are traced with step-level logging for audit and reliability.

Five Best Practices for Secure, Scalable Tool Integration

  1. Start with API Contracts: Define clear inputs, outputs, and error behaviors.
  2. Select the Right Packaging: Choose between OpenAPI and MCP based on integration needs.
  3. Centralize Governance: Use APIM or gateways to enforce consistent authentication and rate limits.
  4. Bind Actions to Identity: Always maintain context of who or what invoked a tool.
  5. Instrument Early: Add tracing and logging at initial stages for easier monitoring and improvement.

What’s Next

The next post in this series will discuss observability for AI agents with Azure AI Foundry and Azure Monitor, including performance tracing and real-time monitoring.

Did you miss the first post? Read about common use cases and design patterns.


References:

This post appeared first on “The Azure Blog”. Read the entire article here