Power BI Semantic Model with Claude in a short demo

Key insights

• How to Power BI published a 45-second Short on Claude and a Power BI semantic model.
• The notes say Claude connects through the Power BI Modeling MCP Server.
• Claude can inspect tables, relationships, and measures, then help create DAX.
• The approach uses Custom Skills to apply naming and modeling rules.
• The model remains the central layer for reusable business logic and governance.

How to Power BI recently published a 45-second YouTube Short about Claude working with a Power BI semantic model. The short itself is very brief, but the accompanying notes describe a workflow where Claude connects through the Power BI Modeling MCP Server and helps with model analysis and changes. The main takeaway is simple: Power BI modeling is being shown as something an AI agent can assist with, not just a manual desktop task.

What's new

The notes describe a connection between Claude and Power BI through the Model Context Protocol, or MCP. In this setup, the Power BI Modeling MCP Server exposes controlled functions that let an agent read model information and, in some cases, carry out modeling changes. The material says Claude can inspect tables, columns, relationships, measures, and model structure.

The same notes also mention Custom Skills for Claude. These are presented as a way to apply company-specific modeling rules, such as naming conventions, DAX standards, and requirements for metadata and roles.

Why it matters

For Power BI makers and analysts, this could reduce the time spent on repeated modeling work. The notes say users can describe tasks in natural language, such as asking for a measure or a modeling change, and Claude can generate an initial response. That may lower the barrier for people who know the business question but not every detail of the Power BI interface.

For admins and data teams, the governance angle matters too. Because the model stays the central layer for business logic, measures and definitions can remain consistent across reports. The notes also frame semantic models as a controlled base for reports, Copilot, and other AI use cases.

How it works

According to the notes, the flow is straightforward:

1. Claude connects to the Power BI Modeling MCP Server.
2. The server exposes a semantic model and the supported actions.
3. Claude reviews model elements such as tables, relationships, and measures.
4. The user asks for a task in natural language.
5. Claude drafts DAX or suggests model changes.
6. The result is reviewed before it is saved or published.

The material also says this can support work that would otherwise happen in Power BI Desktop, PBIP projects, or the Fabric service.

Limitations and open questions

The source material is thin, and the Short description itself does not add detail beyond identifying the video. The notes say changes are reviewed before being saved, which implies human checking still matters, but they do not explain approval flow, permissions, or any licensing scope.

We also could not verify from the material whether this is broadly available, in preview, or limited to certain tenants. The exact depth of Claude's write access versus read access is also not fully explained in the source material.

Sources

• YouTube Short by How to Power BI
• How to Power BI website

Frequently asked questions

What is the main idea in the Short?

It shows Claude working with a Power BI semantic model through the Power BI Modeling MCP Server. The notes say this can help with analysis, DAX, and modeling tasks.

Does Claude fully replace Power BI modeling work?

No. The notes say the output is reviewed before it is saved or published, so human checking still applies.

What is a Power BI semantic model in this context?

The notes describe it as the central reusable data layer for tables, relationships, measures, and business logic. Microsoft now uses the term semantic model instead of dataset.

Is this available for everyone?

The material does not say. It does not confirm licensing, rollout stage, or tenant availability.