Post by @f9finance

I used Claude Code and Microsoft's brand-new Power BI MCP to build a full Power BI dashboard from a raw file, no manual data modeling, no dragging visuals around. This is Power BI plus AI taken to the end of the line. I handed Claude a 4.5 million row point-of-sale file (the kind that won't even open in Excel), a P&L, and my manager commentary, then asked it to be my Power BI developer for a day. It built a clean semantic model, fact and dimension tables, a date table, mapping tables I didn't even ask for, and a folder of measures. Then it planned a four-page report for the coffee shop's leadership team, built every page, and re-themed the whole thing to my brand from a single instruction. If you do FP&A with AI, this is the part that matters: the boring 80% of report building (the setup, the measures, the layout) is the part that eats your day, and it's exactly the part this hands off. I walk through what Microsoft actually shipped (the Power BI MCP server, the report authoring skills, and the Desktop Bridge, all in preview), how to run it with Claude Code instead of GitHub Copilot, and the one save setting (.pbip) that makes any of it possible. This isn't hype. I show where it's genuinely good and where it still makes mistakes, and I end on the real question: would you ship an AI-built report to your CFO yet?