Power BI: Setup MCP for VS Code & Claude

Key insights

• Power BI Modeling MCP: The MCP server lets AI agents act on Power BI models inside editors like VS Code.
  It enables agentic AI to create measures, relationships, hierarchies, and suggest or fix DAX without manual coding.
• Tenant settings: Enable MCP in the Power BI Admin Portal so clients can connect.
  Keep a Power BI Desktop file open and use proper permissions or a service principal for secure access.
• VS Code setup: Install VS Code, add GitHub Copilot Chat (and Copilot) and then install the Power BI Modeling MCP extension.
  Use the pre-release extension only if your scenario requires newer preview features.
• Verify connection: Start Copilot in Agent mode, configure tools to point to the MCP server, and connect to an open PBIX file.
  Test with commands like list tables to confirm the server sees hidden and auto-date tables; accept permission prompts and retry if needed.
• Manual Claude setup: For Claude Desktop, download the extension VSIX, extract it, run the MCP server executable to create configuration, and add that MCP entry to Claude’s developer settings.
  Fully restart Claude after configuring so the MCP server loads correctly.
• Automation & safety: MCP speeds bulk edits, best-practice checks, and documentation while helping with DAX and relationships.
  Always back up models, use tenant governance for security, and restart clients to resolve common connection issues.

Overview of the Video and Its Purpose

The YouTube video from Pragmatic Works demonstrates how to enable and run the Power BI Modeling MCP server to bring agentic AI into the Power BI modeling workflow. In particular, the video shows how to use the MCP server inside Visual Studio Code and how to optionally connect the same server to Claude Desktop for a smoother prompting experience. As a result, viewers learn how an AI assistant can take action on a model to create relationships, improve DAX, add hierarchies, and speed up documentation. Overall, the clip aims to move from setup to practical prompts in a follow-up video, so this installment focuses on configuration and verification.

Core Setup Steps in Visual Studio Code

First, the video walks through administrative prerequisites, starting with enabling the MCP option in the Power BI tenant settings via the Power BI Admin Portal. Next, the presenter installs Visual Studio Code and validates that key extensions are present, notably GitHub Copilot and GitHub Copilot Chat, before adding the Power BI Modeling MCP extension itself. Then, the tutorial points out when you might need the pre-release or beta version of the extension to access the latest features or fixes. This sequence ensures the MCP server can connect to an open Power BI Desktop model and accept agent-driven commands.

After installation, the video shows how to use Copilot in Agent mode and configure tools to allow the extension to act on the model. Viewers see a live verification by connecting Copilot to an open file and running commands such as listing tables, including hidden and auto-date tables. The presenter emphasizes the "Allow" prompts that appear when a connection is made and offers quick tips for troubleshooting failed connects. Consequently, these steps make it possible for AI assistants to translate natural language into the Tabular Object Model or TOM changes.

Testing, Validation, and Practical Commands

Next, the video demonstrates practical prompts that validate the connection and capabilities of the MCP server. For example, the presenter asks Copilot to list tables and to reveal hidden objects, which confirms the server can query metadata and perform read operations on a live model. Viewers also see how Copilot can propose DAX snippets and suggest improvements, which helps speed iterative development and documentation tasks. Therefore, this testing phase serves as a hands-on checkpoint before enabling broader automation.

Moreover, the tutorial highlights how MCP supports bulk operations and validation, so users can apply changes at scale while preserving transaction safety. As a result, teams can perform refactors or naming standardization across many objects without manual edits. However, the presenter cautions that preview features may behave unpredictably, and stresses the importance of backing up models before major changes. Thus, testing small, reversible actions first reduces risk.

Manual Installation for Claude Desktop and Common Gotchas

For users who want to use Claude Desktop, the video covers the manual path of downloading the VSIX from GitHub, extracting the extension, and running the MCP executable to generate a local configuration. Then, viewers update Claude’s developer configuration to point to the local MCP command so the assistant can use the same server. Importantly, the walkthrough avoids posting raw links and instead explains the file locations and command arguments conceptually to help readers reproduce the steps safely. This manual path is useful for people who prefer Claude or need an offline agent integration.

The presenter also explains several common issues: the extension may not appear until Claude is fully restarted, configuration paths must be exact, and permissions on the model or tenant settings can block actions. Additionally, the video shows how to handle “doesn’t show up / doesn’t work” scenarios by fully restarting applications and validating that the MCP server process is running. Consequently, following these troubleshooting steps will address the majority of connection problems during setup.

Tradeoffs, Security, and Operational Challenges

While MCP unlocks powerful automation, it also introduces tradeoffs that teams must weigh carefully. On one hand, allowing an AI agent to modify models reduces repetitive work and accelerates development; on the other hand, it raises governance and security concerns because automated changes can propagate widely and quickly. Therefore, organizations should balance convenience against control by enforcing tenant-level policies, using service principals where appropriate, and requiring backups before applying mass edits.

Furthermore, the technology’s preview status creates operational challenges: features can change, and LLM-driven actions may not always follow intended logic, which means human review remains essential. In addition, complex models with many interdependencies can make automated changes risky, so prudent teams will combine automated steps for routine tasks with manual oversight for strategic changes. Ultimately, this balance between automation and control is key to safely adopting agentic workflows in Power BI environments.

Conclusion and Practical Next Steps

In summary, the Pragmatic Works video offers a practical, step-by-step guide for enabling the Power BI Modeling MCP server in Visual Studio Code and for integrating the same server with Claude Desktop. It walks through installation, verification, and common troubleshooting tips while emphasizing the importance of backups and governance. As a next step, viewers should replicate the setup in a test environment, try simple prompts to validate behavior, and plan governance controls before moving to production. By doing so, teams can explore agentic modeling safely and incrementally while reaping the productivity benefits this integration promises.

Keywords

Power BI MCP setup, Power BI visuals VS Code, Visual Studio Code Power BI extension, Claude AI Power BI integration, build custom Power BI visuals, Power BI developer tools, debug Power BI visuals in VS Code, deploy Power BI visuals