Microsoft Corp. is using its massive Ignite conference in San Francisco today to unveil a wide range of updates to its Foundry artificial intelligence development platform aimed at simplifying how developers build, deploy and manage AI agents.

The previews and general releases are part of the company’s broader effort to streamline agent development and enable enterprises to integrate AI more deeply into business applications. They include expanded support for the Model Context Protocol, agent interoperability, model optimization and a unified context layer designed to give agents a more complete understanding of enterprise data.

Collectively, Microsoft said the enhancements will enable developers to enrich agents with real-time business context, bundle in multimodal capabilities and incorporate custom business logic through a catalog of MCP tools.

MCP catalog

Developers can use the catalog to locate and manage public or private MCP tools from a single interface. It’s also intended to provide deeper integration with business systems. Logic Apps connectors provide access to more than 1,400 systems, including applications from SAP SE, Salesforce Inc., and HubSpot Inc., exposed via MCP. That allows agents to interact with real-time business data and events without requiring custom integration work. The updates also add new tools for transcription, translation, voice and intelligent document processing.

Developers will be able to expose any application programming interface or function as an MCP tool through API management, enabling the reuse of existing business logic.

Foundry now includes a preview of a model router that automatically selects the best model for a given task. It’s designed to optimize performance and cost by analyzing prompts and selecting the most suitable language model. Microsoft said the result is up to 40% faster response times and 50% lower cost. The router currently includes access to 12 models, including the Open AI LLC’s GPT-4.1 and GPT-5,  GPT-oss-120b, DeepSeek-v3.1, Llama-4-Maverick-17B-128E-Instruct-FP8, Llama-33-70B-Instruct and X.ai Corp.’s Grok-4 and Grok-4-fast.

Foundry Agent Service is adding a set of preview features intended to simplify the deployment of hosted agents, built-in memory and support for multi-agent workflows.

Hosted agents allow developers to deploy custom agents without managing containers or infrastructure. Microsoft said the service will provide autoscaling, observability and identity integration to help move prototypes into production. The company noted that developers using frameworks such as Microsoft Agent Framework, LangGraph, CrewAI’s commercial framework or OpenAI Agents Software Development Kit can use the same runtime environment.

Agents in concert

Multi-agent workflows coordinate multiple specialized agents to perform multi-step processes such as onboarding or financial approvals. Microsoft said workflows will support long-running, stateful interaction and include recovery and debugging features.

The memory feature, coming later this year, will enable agents to retain context and conversation history across sessions, providing persistent recall without requiring external data stores.

Microsoft is also introducing a context layer that combines intelligence from Microsoft 365 Copilot with a new semantic layer and Foundry IQ for retireval-augmented generation. The company said its goal is to give agents a unified understanding of user actions, business data definitions and information location.

Fabric IQ, now in preview, extends the semantic layer used in Microsoft Power BI to operational systems, creating what Microsoft calls a “live, connected view” of business entities and relationships. This is meant to allow agents and applications to act on more contextually relevant information. Fabric IQ employs Microsoft’s OneLake data lake, enabling organizations to leverage models across on-premises, hybrid and multicloud environments.

Streamlined RAG

Foundry IQ, also in preview, is a managed knowledge system that supports RAG across multiple data sources, including Azure data services, Microsoft 365 SharePoint, Fabric IQ and the web. Built on Azure AI Search, it automates multimodal data pipelines and performs agentic retrieval operations such as query planning, iterative search, reflection and synthesis. Integration with Microsoft Purview provides for governance and compliance.

The new Foundry Control Plane, now in preview, extends Agent 365 to provide visibility, control and security for agents deployed in the Microsoft cloud. It includes management features such as observability, guardrails and lifecycle tools.

Microsoft said the Control Plane gives teams a unified view of agents operating across Microsoft Foundry, Microsoft Entra, Copilot Studio and external platforms. Observability capabilities including real-time tracing, continuous monitoring, evaluations and red teaming are now available.

Agent control policies, identity and access controls using Microsoft’s Entra Agent ID cloud identity management platform, and integrations with Microsoft Defender and Microsoft Purview are available in preview. The Control Plane also provides cost and usage management through the AI Gateway, enabling centralized usage limits and cost controls for models, agents and MCP tools.

Image: SiliconANGLE/Microsoft Designer