Microsoft has launched Agent Factory in Azure AI Foundry, providing enterprise tools and design patterns for building AI agents with extensible capabilities. The platform supports the Model Context Protocol (MCP) for tool portability and includes built-in connectors to over 1,400 SaaS and on-premises systems through Azure Logic Apps.
Azure AI Foundry addresses enterprise agent development through three capability layers: built-in tools for rapid deployment, custom tools for competitive differentiation, and broad system connectors. Built-in tools include SharePoint and data lake search, Python execution for data analysis, multi-step web research with Bing, and browser automation tasks.
Early enterprise implementations demonstrate significant productivity gains. NTT DATA built agents integrating Microsoft Fabric Data Agent with other enterprise tools, enabling employees across HR and operations to interact with data naturally. The implementation reduced time-to-market by 50% and provided non-technical users with intuitive, self-service access to enterprise intelligence.
Azure AI Foundry includes enterprise-grade security through on-behalf-of authentication for built-in connectors, preserving existing authorisation rules via managed Entra IDs. Custom tools support OpenAPI and MCP integration with managed identities, API keys, or unauthenticated access. Azure API Management provides centralized policy enforcement, authentication, rate limits, and payload validation.
The platform provides step-level logging for every tool invocation, including identity, tool name, inputs, outputs, and outcomes for continuous monitoring and auditing. Microsoft Entra Agent ID integration offers security teams unified directory views of all agents across Microsoft environments.
Organisations gain standardised agent development frameworks with enterprise governance, enabling rapid deployment while maintaining security and compliance. The MCP support provides tool portability across hybrid and multi-cloud environments without vendor lock-in.