For years, automating things inside Microsoft Fabric meant opening the portal, clicking through dozens of menus, and hoping the interface hadn't changed since last week. Microsoft just changed that, at least for Data Agents.
The Fabric Data Agent public API is here. And with it, the ability to create, configure, and publish data agents without ever touching the Fabric portal.
What Is a Fabric Data Agent, Anyway
Before diving into the API, a quick recap of what a Data Agent actually is.
A Fabric Data Agent is an AI-powered feature that lets you ask questions about data stored in OneLake using natural language. No SQL. No DAX. No KQL. You ask which product sold the most in March and the agent generates and runs the query automatically, returning the answer.
It connects to data sources like Power BI semantic models, lakehouses, warehouses, and KQL databases. None of that is new. What changed is how you manage and deploy these agents.
What the Public API Actually Changes
Before this update, creating and configuring a Data Agent was a manual task. You did it inside the Fabric portal or through the SDK, but only from within Fabric notebooks.
With the public API, the Fabric Data Agent SDK now runs on the same REST surface you already use to manage workspaces and other Fabric items. That means the entire lifecycle of an agent, creating, configuring, updating, and publishing, can happen from anywhere:
- Local development environments
- CI/CD pipelines
- Internal portals
- Azure Functions
- Containers
- Any backend service that can make a REST call
The separation between management and runtime is worth understanding. The API handles the management plane: you use it to build and publish the agent. Once published, the agent is consumed through the MCP (Model Context Protocol) endpoint, which is the standard protocol for connecting AI agents to data sources and tools.
Service Principals: The Detail That Matters for Operations
One thing that should not be overlooked: Fabric Data Agents now support service principals.
That means you can authenticate and execute agents using app identities, without relying on user credentials. For anyone building automated pipelines, this is the difference between a solution that keeps running at 3 AM without human intervention and one that breaks every time someone changes their password.
For organizations that need scalable automation with granular access control, this is a meaningful change.
The Connection to Copilot Studio and the MCP Ecosystem
Copilot Studio, in its latest version, now supports MCP servers. That creates a direct bridge between Fabric Data Agents and the agents you build inside Copilot Studio.
In practice: you publish a Data Agent in Fabric, it becomes available via its MCP endpoint, and Copilot Studio can consume it as a data source or tool inside a workflow. No need to duplicate data access logic somewhere else.
The same applies to Microsoft 365 Copilot integrations. Published Fabric data agents show up as available tools in Teams, on the web, and in the M365 desktop app.
Deployment Pipelines and the Dev-to-Production Flow
Another feature that became more relevant with this update: Fabric deployment pipelines now work with Data Agents.
That means you can promote an agent from a development environment to staging, test it, validate behavior against real pre-production data, and only then push to production. The same flow data teams already use for semantic models and dataflows.
For teams that need traceability and change control, this was a missing piece.
Simpler Permissions for Power BI Semantic Models
A practical change that will make life easier for analysts: to connect a Data Agent to a Power BI semantic model, a user now only needs read permission on the model.
Previously, accessing a semantic model through agents or Power BI Copilot required full workspace access or Build permission. That made governance complicated in environments with many users and shared models.
With read permission being enough, it becomes simpler to distribute access to conversational data experiences without creating unnecessary security gaps.
Why This Matters for BI Analysts and Developers
The opening of the Fabric Data Agent public API is not just another item on Microsoft's feature list. It changes how data teams can think about deploying and maintaining conversational agents over their data.
Until recently, building a data agent that worked in a fully automated way, with version control, multi-environment deployment, and service principal authentication, required workarounds or custom solutions. Now it is native to the platform.
For people working with Power BI and Microsoft Fabric day to day, this is worth treating not as a technical curiosity but as a shift in posture. Microsoft is betting that the future of corporate data access runs through AI agents. And it is building the infrastructure to manage those agents with the same rigor applied to any other data artifact.
The SDK is available on PyPI as fabric-data-agent-sdk. Full documentation is on Microsoft Learn. If you already have Fabric workspaces with capacity that supports Data Agents, you can start today.