Testing the new Power BI Modeling MCP Server extension became so easy that everyone can try it without any complex installation steps. https://github.com/microsoft/powerbi-modeling-mcp

Since I read about Palantir a year ago, I was wondering what it's like to chat with AI about your own data and work with it in an agentic way. It's really easy to install the extension for VSCode then opening the folder where your model resides. You will need to use the new file format, pbip and enable pbir for the definitions in the preview settings. You already installed the Github Copilot extension, right?
I like checking the model for technical issues and organizing the objects via tabular editor, this tool will help speeding up that process even if you prefer to make the changes manually. Asking for help related to a specific Power BI semantic model is usually time-consuming as most of the time the answer depends on a technical and business context of your specific model.

I tested this tool with a demo dataset and it gave me great feedback which prompted me to come up with some rules about the suggestions I agree with or prefer to ignore. It's logical to try to build context so that the server could use it for the analysis.

My first use case was checking the naming conventions. For that purpose, we can simply just ask the agent to create a folder named MCP with some txt files by topics such as conventions, guidelines, change log etc. The naming conventions can contain general best practices and company-specific rules and acronyms. Since I had to open and close Power BI desktop to see the changes, I'd prefer to do them manually based on the feedback I get from the MCP server.

We can ask about visuals and colors as well as they are also described in json structure in the power bi project file's json files. A lot of times I missed the difference between black fonts and dark green, now I don't have to worry about that anymore.

We can ask it for ideas for new visuals, of course my expectation is that one out of 10 ideas are actually good for the business concept and the volume and quality of the data.

That's all I wanted to convey for now! Have fun using this tool!