Azure Foundry IQ Adds Serverless Retrieval and MCP

Introduction

Building enterprise AI agents is often less about the model and more about the knowledge layer behind it. Azure Foundry IQ aims to reduce that complexity by giving developers a unified way to connect enterprise content, structured data, and web sources to agent workflows.

Microsoft’s latest announcements matter because they make it easier to move from prototype to production with lower infrastructure overhead, broader data access, and stronger governance.

What’s new in Foundry IQ

Serverless retrieval now in public preview

• Foundry IQ Serverless (Developer tier) is now in public preview.
• Delivers instant retrieval-augmented generation (RAG) without managing clusters or reserved capacity.
• Uses scale-to-zero pricing, which is better suited for bursty agent workloads.
• Billing is planned for late 2026, and customers will not be charged before billing is enabled.

New knowledge sources in preview

Microsoft added more sources that can feed a single multi-source knowledge base, including:

• Work IQ for emails, meetings, files, and Teams messages
• Fabric IQ for data agents and ontology-based business data
• File Search for direct file uploads
• Azure SQL for relational data
• MCP Server for Model Context Protocol knowledge sources

This reduces the need for custom connectors and source-specific retrieval logic.

Web IQ now available in Foundry IQ

• Microsoft Web IQ extends agent grounding to web, news, images, video, and shopping content.
• Microsoft says it honors publisher preferences, offers sub-165 ms latency, and uses zero data retention.
• This gives agents access to fresher external context alongside internal enterprise knowledge.

Knowledge bases are now generally available

Foundry IQ knowledge bases have reached general availability, bringing:

• SLA-backed production support
• Stable APIs
• Compliance certifications
• MCP server support for tools like Claude, ChatGPT, LangChain, and Microsoft Agent Framework
• Network isolation and managed identity support

Retrieval, ingestion, and security improvements

Additional preview updates include:

• Better agentic retrieval quality with fewer token costs
• Layout-aware document ingestion and image enrichment
• Broader SharePoint indexing
• New controls for encryption, permissions sync, and sensitivity-label governance

Why this matters for IT and development teams

For Azure and AI platform teams, these updates lower the operational effort required to deploy grounded agents securely. Instead of stitching together multiple connectors and retrieval systems, teams can centralize knowledge access and expose it through MCP-compatible tools.

For governance and security teams, the new controls help preserve enterprise policy as content flows into AI systems.

Next steps

• Evaluate Foundry IQ Serverless for pilot agent workloads.
• Test the new preview connectors for enterprise and structured data sources.
• Review GA knowledge bases if you are planning production deployments.
• Assess MCP integration options for your preferred agent framework or host.