Microsoft bolsters Fabric with MCP support, OneLake mirroring for Oracle and BigQuery, graph and geospatial tools

Microsoft adds new AI development and OneLake tools to Fabric

Microsoft rolled out a slate of Fabric updates in preview, announced at FabCon Vienna. The headline: easier access to cross-cloud data, agent-friendly development with MCP, plus native graph and geospatial capabilities.

For data and engineering teams, the focus is practical. Less ETL. More direct use of data where it lives. And new primitives for building agents and analytical applications that understand relationships and location.

What's new (preview)

• OneLake mirroring for Oracle and Google BigQuery: Extends Fabric's zero-copy approach to more enterprise datasets. "Zero copy and zero ETL are hot topics right now," said David Menninger of ISG Software Research. "Extending those capabilities to include Oracle and BigQuery will make large amounts of data much more easily accessible in the Microsoft environment, eliminating all the effort to consolidate data and maintain pipelines that perform that consolidation."
• Graph in Fabric: A graph database based on LinkedIn technology that uses neural networks to surface relationships beyond traditional relational models. Common targets: fraud detection, recommendations, and risk analysis.
• Maps in Fabric: Geospatial features that combine streaming location data with mapping and modeling for AI agents and analytics.
• Model Context Protocol (MCP) support: Standardizes how agents connect to enterprise data and external models. Already in Copilot Studio and Azure AI Foundry, now extended to Fabric to keep agent tooling consistent across Microsoft's stack.

"Core to the Fabric vision … is unifying the world's data," said Arun Ulagaratchagan, CVP of Azure Data. "We're methodically going after every major data source with either a shortcut, a connector or mirroring."

Why it matters for data and AI teams

OneLake's broader mirroring support reduces pipeline sprawl, latency, and ongoing maintenance. It also helps keep governance and lineage intact by avoiding copies.

The new graph and geospatial features add native constructs for use cases that are hard to model with rows and columns alone. "The graph database was perhaps the last missing piece in Fabric becoming a complete data platform," Ulagaratchagan said.

William McKnight noted, "The new features in Microsoft Fabric demonstrate a significant update that enhances its capabilities as a comprehensive, enterprise-grade data platform. The additions position Fabric as a key component for customers adopting AI and undertaking data modernization initiatives."

Competitive context

Microsoft's moves track with market direction, but they're not outliers. "Together, these updates represent the breadth and depth of resources Microsoft can apply to this space," Menninger said. "Individually, none of these features is earth-shattering - they didn't solve world hunger - but they've touched on many of the key issues enterprises are facing when working with data."

McKnight added a caution: "While Fabric is investing heavily in AI, it may not yet offer the same level of sophistication or integration of AI directly into analytical workflows as competitors. Databricks has its deep AI/ML capabilities built on an open lakehouse architecture… Snowflake's AI capabilities are maturing to offer an 'easy button' for basic AI use cases."

Practical ways to put this to work

• Consolidate analytics without copies: Mirror Oracle and BigQuery into OneLake and query alongside Microsoft-native data to speed up cross-domain analysis.
• Operational risk and fraud: Stand up Graph in Fabric to detect multi-hop patterns that are invisible to relational joins.
• Location-aware agents: Use Maps in Fabric to feed route, region, or geofence context into AI assistants and monitoring workflows.
• Standardize agent tooling: Adopt MCP across Fabric projects to keep tool calling, data access, and model integrations consistent from dev to prod.

Architecture and governance notes

• ETL reduction: Favor mirroring and shortcuts to minimize duplication, cost, and drift across environments.
• Latency vs. freshness: Validate SLAs per source system; mirroring reduces movement but doesn't remove source-side limits.
• Security: Align Fabric workspaces, data products, and access controls with existing policies; test row/column-level rules on mirrored sets.
• Cost control: Track storage, query, and egress behavior when federating across clouds.

What's next

All new features are in preview, effectively outlining Microsoft's near-term roadmap. McKnight recommends deeper commitment to open standards, more differentiated AI/ML workloads, and finer controls for expert users.

Menninger expects progress on semantics: "Semantic models will be one of the next big frontiers. They provide context to AI and BI. However, they are not easy to create. … I expect we'll see more from Microsoft and others over time."

Resources

• Microsoft Fabric documentation
• Model Context Protocol (MCP)

Skill up

If you're building agentic systems or expanding Fabric skills, browse curated training by vendor and role at Complete AI Training.