Informatica and Microsoft have deepened their collaboration to speed enterprise adoption of agentic AI by connecting Informatica’s Intelligent Data Management Cloud (IDMC) and CLAIRE intelligence directly into Microsoft’s Azure AI Foundry through a new Model Context Protocol (MCP) integration, a library of pre-built GenAI recipes, and expanded analytics interoperability with Microsoft OneLake using the Apache Iceberg table format.
Background
Agentic AI—the class of systems that combine large language models with tool access, memory and decision-making workflows to perform multi-step tasks autonomously—has moved from experimental labs into commercial product roadmaps across cloud vendors and enterprise software firms. Microsoft’s Azure AI Foundry (often shortened to Foundry) and companion services like the Azure AI Agent Service provide a platform for building, hosting and governing agents; the Model Context Protocol (MCP) has emerged as the de facto interop standard for connecting agents to tools and data sources. At the same time, enterprises are demanding that AI systems operate on
trusted, governed data rather than ad-hoc document dumps. That is the space Informatica occupies: cataloging, mastering and assuring data quality across hybrid estates with IDMC and augmenting metadata-driven intelligence via the CLAIRE AI engine. With this announcement the two vendors tie those strengths together: Foundry supplies agent runtime and orchestration; Informatica supplies the data fabric and governance controls that agents should rely on.
What was announced
- MCP integration (Informatica MCP Server for Foundry Agent Service): AI agents created in Azure AI Foundry can connect to Informatica’s IDMC catalog, data quality, cloud data governance and Master Data Management (MDM) services using MCP, enabling near‑real‑time, secure access to governed enterprise datasets.
- GenAI recipes for Foundry: A library of pre-built templates (called GenAI recipes) aimed at accelerating common enterprise use cases — retrieval-augmented generation (RAG) flows, chat-history management, prompt-chaining and industry blueprints such as loan processing and automobile insurance claims workflows. These recipes pair Azure OpenAI and Foundry agent capabilities with Informatica’s metadata and data-management building blocks.
- CLAIRE expansion in Azure regions: Informatica’s CLAIRE AI engine will be natively available in Microsoft Azure regions across the United States and Europe to support regional compliance and data residency requirements when running intelligent data operations.
- OneLake + Apache Iceberg support: Informatica added support for Microsoft OneLake tables backed by the Apache Iceberg table format, improving interoperability for analytics and enabling enterprises to access unified data across diverse sources with industry-standard table formats.
Executives framed the partnership in classic enterprise terms: Informatica’s Chief Product Officer Krish Vitaldevara positioned the work as “a major leap forward” for building compliant, production-grade AI agents, while Microsoft’s Amanda Silver emphasized the centrality of
trusted, governed data to deliver accurate and responsible agentic outcomes. Dayforce was named as an early adopter using the combined stack to create unified customer views.
Why this matters: the enterprise problem Informatica + Microsoft attempt to solve
Enterprises attempting to deploy GenAI in production face three recurring, hard problems:
- Data trust and governance: Models are only as good as the data they use. Unmanaged data leads to hallucinations, regulatory exposures and inconsistent outcomes.
- Operationalizing agents safely: Agents that can execute tools, call external APIs, and access sensitive data increase the attack surface and require robust approval, auditing and control mechanisms.
- Interoperability and scale: Large organizations need agents and analytic workflows to work across on‑prem, cloud and multi‑vendor ecosystems without duplicating data or rebuilding pipelines.
The Informatica–Microsoft expansion directly targets each problem: IDMC provides cataloging, lineage, MDM and quality controls; MCP supplies a common plumbing so agents can call Informatica-hosted tools and datasets; Azure AI Foundry offers an enterprise orchestration plane with approval workflows and runtime controls; and OneLake/Iceberg support reduces data friction between analytics engines. For organizations that have historically separated “AI” from “data ops,” the announcement is an attempt to make those lines disappear.
Technical deep dive
Model Context Protocol integration: how agents will see enterprise data
The Model Context Protocol is an open standard that codifies how model-driven clients (LLMs/agents) can call external tools, query knowledge sources and retrieve structured contextual data. Microsoft’s Azure AI Foundry supports MCP as a mechanism to attach external MCP servers as tools to an agent; the Foundry agent runtime sends tool invocation requests to the MCP server and receives structured responses. Microsoft’s docs make clear Foundry expects MCP servers to be reachable via remote endpoints and surface security guidance for header-based authentication and runtime approvals. Informatica’s new MCP Server for Foundry Agent Service essentially exposes selected IDMC services—catalog, data quality checks, MDM lookups and governed dataset fetches—as MCP tools. An agent running in Foundry can therefore, under policy, resolve a customer record in MDM, verify a dataset’s lineage and quality metrics, and retrieve curated content rather than raw, ungoverned files. Because MCP supports structured tool outputs and conversational context threading, agents can use that structured context to improve answer accuracy and to attach provenance metadata for auditing. Informatica describes this as
near‑real‑time connectivity to governed datasets. Key technical notes and caveats:
- Azure AI Foundry’s MCP tooling currently requires remote-accessible MCP server endpoints (local-only MCP servers must be hosted in cloud container services to be reachable). Authentication headers and tokens can be provided per-run (they are not persisted by Foundry), which helps reduce long-lived credential risk but requires careful orchestration.
- Using an external MCP server means enterprises must evaluate non‑Microsoft retention, location and logging practices for any data routed through that server; Microsoft explicitly recommends auditing and careful review.
GenAI recipes: speed vs. customization
The GenAI recipes are pre-engineered templates that encapsulate common patterns—RAG pipelines with retrieval connectors, chat history management, prompt-chaining for multi-step workflows, and agent orchestration blueprints using Foundry runtime capabilities. These recipes aim to accelerate proof-of-value projects and to provide consistent, repeatable patterns that embed governance checks.
Benefits:
- Reduces time-to-prototype by reusing tested architectures for RAG, memory and tool chaining.
- Embeds data-management touchpoints (catalog lookups, data quality gates, MDM checks) into application flows.
- Includes industry-specific scaffolds (e.g., loan processing, insurance claims) so vertical teams can start from a compliance-aware baseline.
Trade-offs and realities:
- Recipes simplify common paths, but production readiness still requires customization—not least for regulatory needs, model selection, latency tuning and security hardening.
- Enterprises should treat recipes as blueprints, not turnkey applications—especially when integrating with internal systems where business logic and edge cases vary widely.
CLAIRE in Azure regions and the data residency angle
Informatica’s CLAIRE AI engine is a metadata-driven intelligence layer that powers search, reasoning over lineage and automated pattern detection inside IDMC. Making CLAIRE
natively available in Azure regions in the United States and Europe addresses a practical requirement: many customers require that AI processing touching personal, financial or regulated data occur within specific geographic boundaries for compliance. Running CLAIRE components inside regionally appropriate Azure infrastructure helps reduce cross-border data movement, which simplifies compliance with data‑locality rules. Operational implications:
- Customers will still need to validate whether metadata and derivatives generated by CLAIRE are themselves subject to residency rules.
- Legal/compliance teams should work with cloud region mappings and process logs to ensure that audit trails show local processing when required.
OneLake + Apache Iceberg: reducing analytics friction
Support for Microsoft OneLake tables backed by Apache Iceberg means Informatica can read and write table formats compatible with a broad ecosystem (Snowflake, Spark engines, Presto/Trino, and Fabric), enabling "single-copy" workflows where analytics and agent contexts do not need duplicated data extracts. OneLake can virtualize Iceberg and Delta metadata so different engines can see the same logical dataset through their native metadata lens. This supports large-scale analytics and helps agents performing RAG or structured queries to use up-to-date data without extra ETL steps. Practical notes from the field:
- Some OneLake/Iceberg features are preview and have operational constraints—teams must verify the specific preview capabilities (e.g., direct Iceberg writes vs. virtualization) against their workloads. Community and Microsoft docs show the virtualization model is evolving.
Benefits for enterprise IT and data teams
- Faster time-to-value for agentic projects: Pre-built recipes and an MCP bridge shorten the integration loop between agents and governed data stores.
- Improved model reliability: Agents can fetch curated, quality-marked context rather than aggregating raw documents, lowering the rate of hallucinations and increasing verifiable accuracy.
- Clearer audit trails: Tying agent data accesses to IDMC catalog entries and MDM lookups produces provenance that auditors and regulators can inspect.
- Interoperability across analytics platforms: Iceberg support and OneLake integration reduce data duplication and permit analytics engines and agent pipelines to operate on a single logical dataset.
- Regional compliance controls: CLAIRE availability in regional Azure zones helps address data residency and sovereignty constraints for EU and US customers.
Risks, gaps and governance red flags
The announcement is strategically sensible—but it isn’t a panacea. The largest enterprise risks fall into five categories:
- MCP surfaces new attack vectors. MCP enables programmatic tool invocation and can pass tokens or credentials at runtime; if a compromised agent or client can leverage MCP calls, the attack surface widens. Microsoft and third-party reporting have highlighted risks like token exfiltration and prompt injection when tool access is enabled without strong controls. Enterprises must implement approval workflows, token expiration, and robust runtime monitoring.
- Data leakage through tool returns and memory. Agents that retrieve snippets or structured responses can inadvertently expose sensitive columns or PII unless responses are filtered, redacted or masked. RAG patterns must be paired with content filters, entity masking, and strict data-scoping policies at the MCP layer.
- Operational complexity and complacency. Pre-built recipes can create a false sense of security—teams may underestimate the need for stress tests, red-team exercises and model evaluation metrics in production. Recipes are a starting point; production-grade deployments demand bespoke instrumentation and SLOs.
- Vendor and metadata coupling. Deep integrations between Foundry and IDMC are valuable, but they can tighten coupling between cloud and data-management vendors. Organizations with multi-cloud or hybrid strategies should design abstractions so migration or vendor shifts remain possible without wholesale re-engineering.
- Preview features and changing APIs. Several OneLake/Iceberg behaviors and Foundry MCP capabilities are evolving. Using preview features in critical production flows requires contingency planning for API changes, latency differences and support model limits.
Practical guidance: hardening agent deployments that use Informatica + Foundry
- Implement least privilege for MCP tools and require per-run approval flows. Use Foundry’s tool approval workflow and restrict which agents can call which MCP servers.
- Place sensitive datasets behind a data-masking and tokenization layer in IDMC; expose only the sanctioned, reduced view to MCP consumers. Use MDM to deliver canonical keys rather than full records.
- Enforce contextual logging and immutable audit trails for every agent invocation and dataset fetch. Store logs in a hardened SIEM and instrument automated anomaly detection.
- Red-team your RAG and prompt-chaining flows to find ways an agent might formulate unsafe or unauthorized tool calls. Simulate malicious prompts and unavailable services to ensure graceful failure modes.
- Validate compliance with regional rules: confirm that CLAIRE processing nodes are located in the required Azure regions and document metadata processing residency for regulators.
- Use formal model evaluation metrics (accuracy, calibration, provenance coverage) for every recipe you launch into production; track drift and retrain or reconfigure retrieval indices when coverage degrades.
What this means for different stakeholders
CIOs and data leaders
This integration lowers the operational bar for using governed data with agentic applications. It enables a governed path to production while concentrating controls in IDMC and Azure’s management plane. However, leaders must approve investment in observability, incident response and compliance teams to manage the new agent footprints.
Security and compliance teams
MCP and agent runtimes create novel policy enforcement points. Security teams should require pre-deployment threat models for any agent that will access regulated data, enforce token and session controls, and mandate on-call protocols for agent misbehavior or data-exfiltration detection.
Data engineers and platform teams
Expect new work: enabling MCP endpoints, customizing GenAI recipes for internal schemas, mapping OneLake/Iceberg semantics to on‑prem sources, and instrumenting lineage and SLOs. Much of the heavy lifting will be integration and testing.
Business teams and product owners
The recipes and agent patterns offer a rapid prototyping path to customer‑facing productivity gains (claims automation, document triage, intelligent case routing). But business owners must own outcome metrics and be prepared to pause or roll back features if model decisions degrade customer trust.
Industry perspective and ecosystem implications
Multiple vendors and cloud partners are converging on open standards like MCP and Iceberg to reduce point integrations and to accelerate agent innovation. Microsoft’s Foundry investment and the wider push for MCP adoption (including desktop and OS integration scenarios) indicate a future where agents are first-class enterprise workloads. Integrations like Informatica’s help shift the conversation from isolated LLM experiments to governed, auditable agent systems that fit into existing data‑management processes. However, the ecosystem is changing rapidly: APIs evolve, previews become GA on uneven timetables and third-party tools must continuously validate assumptions about metadata virtualization and table format support. Firms that adopt too early without clear governance discipline risk operational surprises and compliance headaches.
Bottom line
The Informatica–Microsoft expansion is a pragmatic and consequential move: it stitches a market-leading data-management fabric to an enterprise agent runtime, closing a glaring gap between
what agents want to use (model context and tools) and
what enterprises must protect (trusted, governed data). The technical integration—MCP server, GenAI recipes, CLAIRE region expansion and OneLake/Iceberg support—creates a meaningful path to build agentic applications that are more auditable and aligned with regulatory needs. That said, the integration is not a turnkey safety guarantee. Successful, secure agent deployments will require disciplined governance, security engineering, red‑teaming, and careful handling of preview features. Organizations should treat the new capabilities as powerful accelerants—provided they pair them with robust operational guardrails.
Action checklist for teams ready to evaluate or adopt
- Inventory sensitive datasets and tag them in IDMC before enabling MCP access.
- Run a pilot using a GenAI recipe for a limited, low‑risk use case (e.g., internal document retrieval) and measure hallucination and provenance metrics.
- Implement runtime approvals and per-run auth tokens for MCP tool calls; validate that headers are never persisted in your environment.
- Configure CLAIRE processing to run in the Azure region(s) that meet your data residency requirements.
- Test OneLake/Iceberg read/write behavior in a sandbox to understand virtualization limits and performance characteristics.
- Define SLAs, on‑call procedures, and rollback plans for agent-driven workflows.
The partnership represents a maturing phase in enterprise GenAI: vendors are no longer offering isolated model endpoints or point‑solutions, but integrated stacks that try to reconcile the speed of agentic automation with the demands of governed, production data operations. For organizations that invest in the people, process and tooling to manage the complexity, Informatica and Microsoft’s combined offering provides a pragmatic foundation to scale agentic AI across regulated and mission‑critical environments.
Source: IT Brief UK
Informatica, Microsoft expand partnership for agentic AI