How to Connect Claude to Power BI for End‑User AI Analytics

Power BI dashboards are great at showing what happened. But when stakeholders start asking why it happened and what to do next, traditional reports quickly hit their limits. That’s where combining Power BI with Anthropic’s Claude and Microsoft Fabric really changes the game.

By connecting Claude directly to your Power BI semantic models via MCP (Model Context Protocol), end users can simply chat with their data: ask why a month overperformed, compare stores, or request improvement ideas—without touching a single visual.

Why Claude + Power BI Is a Big Deal for Analytics

Most Power BI reports are built around visuals and KPIs: charts, slicers, and a handful of key metrics. They’re great for monitoring performance, but they rarely answer deeper questions like:

Why did June outperform every other month?

Why was March so weak?

Which products, stores, or regions actually drove those results?

With Claude connected to your Power BI semantic model, an end user can simply type a question like:

“Explain why June had the most sales and why March had the fewest. What were the key drivers and executive-level insights?”

Claude then queries the underlying model (not just what’s visible on the report) and returns:

• Breakdowns by store, region, product, and month

• Comparisons of transaction volume, units sold, and net sales

• Executive-style summaries of what drove performance up or down

This goes beyond what’s on the page. Claude can surface dimensions and metrics that exist in the model but aren’t shown in the report, effectively turning your semantic model into an interactive, AI-powered analyst.

From Dashboards to Conversations With Your Data

Once the connection is set up, the end-user experience is straightforward. They don’t need to know anything about MCP, Fabric, or connection strings—they just talk to Claude.

For example, imagine you see that Store Fama is performing well and Store Palma is lagging. Instead of building a custom comparison page in Power BI, you can ask:

“Compare Fama to Palma and explain the differences in detail. How can I improve the store performing poorly?”

Claude will:

• Compare both stores across net sales, transactions, units sold, and product categories

• Highlight which products are dragging performance down or leading the pack

• Analyze performance over time (e.g., by month)

• Offer concrete suggestions for improving the weaker store based on the data

Traditional reports can show you the numbers, but they rarely tell you what to do. Claude bridges that gap by turning descriptive analytics into diagnostic and prescriptive insights—all driven by your existing Power BI model.

If you want a deeper understanding of how Claude uses MCP, skills, and structured tools under the hood, check out this in-depth guide: The Ultimate Claude Code Guide: Models, MCP, Skills, Sub‑Agents & Memory.

What You Need Before Connecting Claude to Power BI

To make this integration work, a few prerequisites must be in place on the Power BI / Fabric side.

1. A Premium Workspace With XMLA Endpoint

The key enabler is the XMLA endpoint, which lets external tools (including LLMs) connect to your semantic models.

You’ll need:

• A workspace hosted on Power BI Premium or Fabric capacity (Premium per user, full Premium capacity, or Fabric equivalent)

• A published semantic model (dataset) in that workspace

In the workspace settings, the type should clearly indicate Premium or Premium per user. That’s what unlocks the XMLA endpoint.

Inside the semantic model’s settings, you’ll see a Server settings section with a connection string. This is the XMLA endpoint URL that Claude will use via MCP. If you don’t see this, your workspace is not on Premium capacity.

Optionally, in tenant settings, you can enable features like “Allow XMLA endpoints and Analyze in Excel with on-premise semantic models.” While not always strictly required, it’s commonly enabled in setups that use external tools.

2. MCP Server Installed and Running

On the Claude side, everything hinges on MCP (Model Context Protocol), which lets Claude talk to external systems like databases, APIs, and in this case, Power BI semantic models.

Each user who wants to connect Claude to Power BI will need to:

• Install the Power BI Modeling MCP server on their machine (a quick one-time setup)

• Ensure it’s running and visible in Claude’s settings under the Developer section (e.g., “Power BI modeling MCP” listed as active)

Once that’s in place, Claude can use the MCP server to call into Microsoft Fabric and authenticate the user against their own Power BI environment.

For a broader look at using MCP to turn models into powerful agents, you might also like How to Turn Local LLMs into Powerful AI Agents with Ollama and MCP.

How to Connect Claude to a Power BI Semantic Model

With the prerequisites ready (Premium workspace, semantic model, MCP server installed), the actual connection steps are surprisingly simple.

Step 1: Grab the Fabric Connection String

In Power BI / Fabric:

• Open your workspace

• Go to the semantic model (dataset) you want Claude to use

• Open its Settings

• Scroll down to Server settings and copy the connection string (XMLA endpoint)

This string is what Claude will use to connect through the MCP server.

Step 2: Use the “fabric connect” Command in Claude

In Claude:

• Start a new chat

• Type a command in this pattern:

fabric connect <your XMLA endpoint connection string>

For example:

fabric connect powerbi://api.powerbi.com/v1.0/myorg/YourWorkspaceName

Claude will then:

• Use the Power BI MCP server

• Call into Microsoft Fabric

• Prompt you to authenticate with your Power BI account (if needed)

Once authenticated, Claude is now connected to that semantic model and can start querying it.

Step 3: Save the Connection as a Reusable Skill

To make this easier for non-technical end users, you can wrap the connection in a simple, reusable skill.

After running the fabric connect command, ask Claude to:

“Save this connection as MS Sales Report. If users say ‘connect to MS Sales Report’, execute the string above. Create a skill for this and allow me to share that skill.”

Claude will:

• Create a named skill (e.g., “MS Sales Report”)

• Associate it with the Fabric connection

• Optionally let you download or share the skill definition

Now, instead of pasting connection strings, end users can simply say:

“Connect to the MS Sales Report.”

Claude will run the underlying connection logic and attach itself to the right semantic model in the background.

Scaling Analytics to Every End User

Once this pattern is in place, you can think of your analytics stack in three layers:

• Semantic model: Carefully modeled data in Power BI / Fabric (measures, relationships, dimensions)

• Reports: Visual dashboards for quick monitoring and descriptive insights

• Claude: An AI analyst that sits on top, answering why and how questions and suggesting what to do next

Instead of endlessly adding new report pages and slicers for every stakeholder request, you can:

• Invest in robust semantic modeling

• Publish clean, focused dashboards

• Let end users explore deeper through natural language with Claude

This approach scales analytics in a way that’s both powerful and user-friendly. Business users get executive-level explanations and recommendations without needing to learn DAX, build visuals, or understand the underlying data structure.

As LLMs like Claude continue to improve and MCP integrations expand, this kind of “chat with your BI” setup is likely to become a standard part of modern analytics stacks.