Skip to main content
cancel
Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for 
Search instead for 
Did you mean: 

Did you hear? There's a new SQL AI Developer certification (DP-800). Start preparing now and be one of the first to get certified. Register now

tompeplow
‎04-16-2026 10:00 AM

Give your AI agent the keys to OneLake: OneLake MCP (Generally Available)

Have you ever tried to understand what's stored in your Fabric items? Would you even know where to begin? I had 92,000 UK property transactions sitting in an open mirrored database. Rather than spending hours sorting through documentation, I just asked my AI agent:

"Document what's in the House Price Open Mirror in my UK Property Data workspace."

One prompt. No code. No clicks. The agent found the workspace, discovered the table schema, mapped the storage structure, and assessed the health of the database — all on its own. The following is a condensed version of what the agent produced:

Screenshot_of_an_AI_agent_response_documenting_a_mirrored_database_in_OneLake_shScreenshot_of_an_AI_agent_response_documenting_a_mirrored_database_in_OneLake_sh

Figure: An AI agent documents an open mirrored database from a single natural-language prompt — no code, no clicks.

The agent navigated from workspace → item → table schema → physical files, switching between table APIs and file APIs as it went. It read monitoring files to check replication health. It measured file sizes to assess optimization. All through a single natural-language conversation.

OneLake tools in the Fabric MCP server make the previous example possible.

How does it work?

MCP is the language AI uses to talk to the tools it needs to get things done. OneLake's promise is all your data, in one place — shortcuts, mirroring, and the catalog connect data from across your estate into a single logical lake. The MCP tools let an AI agent work with all of it through conversation.

Fabric items store their data in OneLake in open formats — from lakehouses to mirrored databases, and even KQL databases and semantic models with OneLake availability enabled. That makes it a natural fit for MCP: a single set of tools can explore them all.

The OneLake tools connect to your tenant's data plane, operating under your existing Azure identity and Fabric permissions — your AI agent can only access what you can access. The tools cover file system APIs for browsing, reading, and writing files, table APIs for discovering schemas and table metadata, and workspace and item discovery to help agents orient themselves — 19 commands in total.

To get started, refer to the Use AI agents with OneLake through MCP, and for the full command reference, refer to the OneLake tools README documentation.

If you manage data in Fabric today, the MCP tools let you do it through conversation instead of clicking through the portal. If you're a developer, they let you automate it. An admin could ask an agent to inventory every item in a workspace. A data engineer could check table optimization across lakehouses. An analyst could explore an unfamiliar dataset without writing a query.

How the example works

In the previous example, the database was built using the OneLake API deep dive — a .NET app that streams UK house price data into an open mirrored database via the Blob Storage API. Fabric converts the raw Parquet uploads into managed Delta Lake tables automatically.

The MCP tools let the agent inspect both sides — the landing zone where data arrives and the optimized tables that Fabric produces. It uses table APIs to understand the schema and file APIs to map the physical storage. Because OneLake stores everything in open formats, the agent can read table metadata without running queries and browse files without opening a notebook.

Now imagine scaling this up. You could ask an agent to scan every item in a workspace — or across workspaces — and document what's there, how big it is, and how healthy it looks. That's a lot of items to click through in the portal. With MCP, it's one conversation.

Get started

The OneLake tools are included in the Fabric MCP server VS Code extension. If you already have the extension installed, you'll get the OneLake tools automatically.

The tools use Azure authentication — sign in with az login and make sure you have access to the Fabric workspaces you want to explore. The tools operate under your existing Fabric permissions, so your agent can only see and modify what you can — no additional roles are required beyond your normal workspace access.

For manual configuration or use outside VS Code, refer to the setup instructions in the Fabric MCP Server README.

What's next

The OneLake tools are part of the broader Microsoft MCP initiative. They're open source, extensible, and we're actively developing them. We're already thinking about what comes next — tools for setting up and managing shortcuts, configuring OneLake security policies, and enabling OneLake diagnostics to help you monitor and troubleshoot your data estate.

We'd love to hear from you. What scenarios would you use these tools for? What's missing? If you've built something interesting with the OneLake MCP tools — an automated documentation pipeline, a governance agent, and a migration workflow — share it with us. Open an issue or discussion on the GitHub repository or share your examples with the community. Your feedback and real-world use cases will shape what we build next.

Learn more

0