Power BI MCP ServerThe Complete Setup Guide for AI-Powered Semantic Modelling

What Is the Power BI MCP Server?

The Model Context Protocol (MCP) architecture consists of three components: the Host (such as VS Code or Claude Desktop), the Client (such as GitHub Copilot or Claude), and the Server (in this case, the Power BI MCP server). Think of MCP as a standardised “Type-C port” for AI: a universal interface that plugs AI into external systems without custom integrations.

Instead of building bespoke connectors for every tool, MCP provides a common protocol that any AI assistant can use to interact with Power BI semantic models. Much like how generative AI has transformed accounting reporting, Microsoft's implementation includes two distinct Power BI MCP servers, each serving different purposes.

Local vs Remote: The Two Power BI MCP Servers

Setup Path 1: VS Code + GitHub Copilot [RECOMMENDED]

This is the recommended path for developers building models on their local machines using Power BI Desktop.

Installation Steps

Setup Path 2: Claude Desktop (Local Server)

Claude Desktop supports MCP natively and works exceptionally well with the local Modeling Server. Perfect if you prefer Claude 3.5 Sonnet over GitHub Copilot.

1. Install VS Code and the Power BI Modeling MCP extension first (this installs the required executable).
2. Open File Explorer to: %USERPROFILE%\.vscode\extensions\. Find the analysis-services.powerbi-modeling-mcp folder. Go into the server subfolder. Right-click powerbi-modeling-mcp.exe and select Copy as Path.
3. Install Claude Desktop from claude.ai/download.
4. Open Claude Desktop. Go to: File → Settings → Developer → Edit Config.
5. Add the following JSON configuration (replace the path with your actual exe path, using double backslashes):

{
  "mcpServers": {
    "powerbi-modeling-mcp": {
      "command": "C:\\Users\\YourUsername\\.vscode\\extensions\\analysis-services.powerbi-modeling-mcp-0.1.9-win32-x64\\server\\powerbi-modeling-mcp.exe",
      "args": ["--start"]
    }
  }
}

6. Completely quit Claude Desktop and restart it.
7. Look for the hammer icon in the chat input to confirm the server is active. Open Power BI Desktop, and start prompting!

Setup Path 3: Remote Power BI MCP Server (Cloud)

For users with models published to Microsoft Fabric or Power BI Service. Requires a Fabric capacity (F2+) or Power BI Premium Per User (PPU) licence.

{
  "servers": {
    "powerbi-remote": {
      "type": "http",
      "url": "https://api.fabric.microsoft.com/v1/mcp/powerbi"
    }
  }
}

Practical Use Cases and Prompts

Important Limitations and Security

• No Visual Edits: The server cannot modify report pages, visuals, or diagram layouts. It focuses exclusively on data modelling.
• Token Limits: A medium-sized model.bim file (~26k lines of JSON) requires around 210,000 tokens. Scope your requests to specific tables or measure groups on large models, similar to how you would structure MACRS depreciation schedules, to avoid context degradation.

Recommended YouTube Resources

Conclusion

Power BI is becoming conversational. MCP bridges generative AI and analytical BI, removing the friction between natural language and semantic modelling. You now have two clear paths: the local Modeling MCP for developers building models (works with free Power BI Desktop), and the remote MCP for cloud-based data querying (requires Fabric or PPU). As you scale these automated workflows, you can replicate complex reporting environments directly within FinDataPro's platform.

Start with the local Modeling MCP if you use Power BI Desktop regularly. It requires no additional licence and integrates directly with VS Code or Claude Desktop. Explore the official Microsoft documentation at Microsoft Learn and the GitHub repository at GitHub. This workflow shift is worth adopting now, before it reaches General Availability, fundamentally replacing manual DAX writing and tedious model documentation with AI-assisted efficiency.

Prashant Panchal• ACA | FMVA® | 19 Years in Finance

Prashant Panchal is a Chartered Accountant (ACA) and Financial Modelling & Valuation Analyst (FMVA®) with 19 years of experience in finance, FP&A, and financial modelling across the GCC region. He is the founder of FinDataPro.