Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
The Model Context Protocol (MCP) is an emerging standard in the AI landscape that allows AI systems to connect with tools and data outside of themselves. It defines how an AI model can discover what’s available and interact with it in a consistent way. Instead of building one-off integrations, MCP offers a standard way to plug things in that works across different apps and services. This makes it much easier for AI systems to go beyond their built-in knowledge while keeping things consistent. It also helps teams move faster, since they don’t have to reinvent the same connections every time.
MCP has two main parts: the client and the server.
An MCP client is the app or experience the user interacts with. It’s where you ask questions or trigger actions. The client reaches out to MCP servers to find tools and use them. For example, Visual Studio Code can act as an MCP client when it connects to external tools to retrieve data, or help you write and run code.
An MCP server exposes tools, data, or services so they can be used by clients. It tells the client what’s available and how to use it. For example, a Fabric data agent can act as an MCP server by exposing enterprise data and queries that an AI system can use.
Together, the client and server make it easy to connect AI systems with real data and actions, without building custom integrations every time.
Important
This feature is in preview.
Prerequisites
- A paid F2 or higher Fabric capacity, or a Power BI Premium per capacity (P1 or higher) capacity with Microsoft Fabric enabled.
- Fabric data agent tenant settings is enabled, including the Capacities can be designated as Fabric Copilot capacities setting.
- Cross-geo processing for AI is enabled.
- Cross-geo storing for AI is enabled.
- At least one of these, with data: A warehouse, a lakehouse, one or more Power BI semantic models, a KQL database, or an ontology.
- Power BI semantic models via XMLA endpoints tenant switch is enabled for Power BI semantic model data sources.
- For Power BI semantic models used with a data agent, ensure users who interact via the agent have Read permission on the semantic model. Workspace Member or Build permission isn't required for interaction.
How it works
Fabric data agents can also function as MCP servers. When used as an MCP server, a Fabric data agent exposes a single tool. This tool represents the data agent itself, so external AI systems can interact with it through the MCP protocol. Because of this, it's important to provide a detailed and accurate description when publishing a Fabric data agent. The description becomes the tool description for the MCP server. External AI systems use this description to determine when and how to invoke the data agent. A clear and comprehensive description ensures that the agent is correctly understood and can be effectively used in AI workflows.
The Fabric data agent as an MCP server is valuable for people who build or test their own AI systems. It allows them to connect directly to the data agent and access enterprise data that lives in Fabric OneLake without having to switch between different tools or platforms. This capability makes it easier to integrate enterprise knowledge into AI experiments and development workflows, all within a single environment.
Note
Currently, you can use the Fabric data agent MCP server only in VS Code. If you’re using your own MCP client, it can also work, as long as you set up authentication
To get started, after publishing the data agent, go to the Settings of the agent and open the Model Context Protocol tab. Here you see the following information:
- Data agent MCP server name
- MCP server URL
- Data agent MCP tool name
- MCP server tool description
You can also download the mcp.json file from this tab. Use this file to configure the MCP server in VS Code.
Setting up the MCP server in VS Code
Open VS Code and select a folder to work in.
Inside this folder, create a folder named .vscode.
Inside the inner folder, create a file called
mcp.json.VS Code automatically displays a blue Add Server button at the bottom right of the window.
Select Add Server and select HTTP. You're prompted to enter a URL. You can copy the MCP server URL from the Setting tab of the data agent as was shown earlier.
Press Enter and provide a name for your MCP server. Use this name to display the data agent MCP server in your VS Code environment.
VS Code attempts to authenticate with the server. Select Allow and sign in with your credentials.
Enabling Agent Mode
After adding the MCP server, enable Agent Mode in VS Code. Agent Mode lets VS Code act as an orchestrator interface, connecting your editor with MCP servers to interact with external tools like the Fabric data agent. To enable it:
In VS Code, go to the Command Palette (Ctrl+Shift+P or Cmd+Shift+P).
Search for Enable Agent Mode and select it.
Confirm any prompts to activate the mode.
When Agent Mode is active, select an orchestrator to handle your queries. Available orchestrators in public preview include GPT-5, GPT-4.1, Claude Sonnet 4.5, Gemini 2.5 pro, and many more. The orchestrator manages the flow of information between your queries in VS Code and the Fabric data agent MCP server.
Using the Fabric Data Agent MCP Server
When you enable Agent Mode and select the orchestrator:
- You can start asking questions directly from VS Code.
- The orchestrator routes your queries to the Fabric data agent MCP server.
- The agent returns answers based on the knowledge it has access to, including organizational data stored in Fabric OneLake.
By functioning as an MCP server, the Fabric data agent allows users to integrate organizational knowledge into AI workflows, perform experiments, and develop AI solutions without leaving VS Code. This integration streamlines access to OneLake data and enhances productivity for developers and business users alike.