AI agents are increasingly becoming part of how developers work with data. But generating SQL is only one step in a typical warehousing workflow. Users still need to understand the right objects, run and validate queries, troubleshoot failures, and investigate performance when something goes wrong.
Today, agents often either lack the access or the product-specific context needed to help across that full workflow.
Model Context Protocol, or MCP, gives agents a standard way to connect to external systems and invoke tools. Agent Skills complement MCP by giving agents reusable, task-specific instructions that guide how they approach and complete work.
For Fabric Data Warehouse, we are bringing both together. The Fabric Data Warehouse MCP Server (remote), now in preview, gives agents a governed way to execute SQL queries against the warehouse, while Skills for Fabric give them the Fabric-specific knowledge to use that access effectively.
The result is a more connected workflow, where an agent can help users move from writing SQL, to validating it against live data, to troubleshooting and monitoring the warehouse without treating each step as a separate experience.
Figure: Fabric Warehouse MCP Server & Skills for Fabric architecture
Fabric Data Warehouse MCP Server (remote)
For an agent to help beyond drafting SQL, it needs a way to work with the actual warehouse.
The Fabric Data Warehouse MCP server (remote), now in preview, gives MCP-compatible agents and clients a managed way to connect to Fabric Warehouses and SQL analytics endpoints.
Agents can connect in two ways:
- Global endpoint gives the agent flexibility to work across Fabric Warehouses by providing the workspace and item context as part of the interaction.
https://api.fabric.microsoft.com/v1/mcp/dataPlane/sqlEndpoint
- Item-scoped endpoint binds the connection directly to a specific Fabric Warehouse.
https://api.fabric.microsoft.com/v1/mcp/dataPlane/workspaces/<workspace-id>/items/<item-id>/sqlEndpoint
Authentication uses Microsoft Entra ID OAuth 2.0, and the user’s existing Fabric and SQL permissions continue to apply. Teams do not need to create a separate security model for agent access, and users do not need to install or maintain a local MCP server.
The server exposes an execute_query tool, which lets an agent execute T-SQL against a Fabric Warehouse. LLMs are increasingly good at generating T-SQL, but the real value comes from being able to act on it. With this MCP server, the agent can run T-SQL against live warehouse data, respond to errors, and continue iterating from the result.
For setup, endpoint configuration, permissions, and examples, explore the Fabric Data Warehouse MCP server documentation.
Skills for Fabric Data Warehouse
Access to the warehouse solves one part of the problem. The agent still needs to understand how Fabric Data Warehouse works.
Skills for Fabric are Agent Skills that provide AI coding tools with Fabric-specific instructions, including supported patterns, commands, query syntax, and recommended approaches for common tasks.
Skills do not execute code themselves. Instead, the AI tool or agent uses the skill as guidance and calls the appropriate Fabric interface, including the remote MCP server for SQL execution.
For Data Warehouse, the skills cover three areas:
- Authoring: Creating and modifying warehouse objects and data, including DDL and DML operations.
- Consumption: Querying and exploring warehouse data, discovering schemas and columns, generating T-SQL, and working with results.
- Operations: Performance and diagnostics, including investigating slow queries, SQL pool pressure, workload behavior, and other performance issues using queryinsights.
Instead of requiring the user to know which Fabric-specific syntax, system view, or diagnostic path to use, the skill gives the agent that context as part of the workflow.
The Data Warehouse skills are open source as part of the Skills for Fabric repository. You can also learn more about how skills work in the Skills for Fabric documentation.
Data Warehouse MCP server and Skills for Fabric in practice
Together, Skills for Fabric and the MCP server let agents work through multi-step warehouse tasks, not just generate a single query. Skills guide the approach, while the MCP server provides the execution path and live results needed to continue the workflow.
For authoring, imagine a user asks:
“Create a sales table for these incoming files, load the data, and make sure everything landed correctly.”
The agent can use the authoring skill to determine the appropriate Fabric pattern, inspect the existing schema, generate the required CREATE TABLE and COPY INTO statements, and execute them through the MCP server. It can then validate the load by checking row counts and sampling the data. If something fails, the agent can use the error returned from execution to revise the SQL and continue.
For consumption, a user might ask:
“Revenue in the West region looks lower this quarter. Help me understand why.”
The agent can use the consumption skill to identify the relevant warehouse objects and begin with a regional revenue comparison. Based on the results, it can drill further into dimensions such as product category, channel, or time period, running additional queries through the MCP server as the investigation develops.
For operations, consider:
“Our warehouse has been slower since this morning. What changed?”
The operations skill can guide the agent through the appropriate queryinsights views and diagnostic steps. Using the MCP server, the agent can identify slow or high-CPU queries, check for SQL pool pressure, review execution patterns, and investigate other workload signals before summarizing what it found.
Across these scenarios, the user starts with the outcome they want. Skills for Fabric provide the Fabric-specific guidance for working through the task, while the MCP server gives the agent a way to execute each step and use the results to determine what comes next.
Get Started
Use the Fabric Data Warehouse MCP server documentation to connect an MCP-compatible agent to a Warehouse or SQL analytics endpoint.
Then explore the Skills for Fabric documentation and the Skills for Fabric repository to add Fabric Warehousing specific guidance for authoring, consumption, and operations to your agent.
Together, remote MCP and Skills for Fabric make it easier for agents to support warehouse development end-to-end, from building and querying data to troubleshooting and operating the warehouse.
Microsoft Employee