Power BI 'MCP Server' adds DAX AI agents to assist with modeling and queries
- Microsoft’s Power BI team put its Modeling MCP Server and Remote MCP Server into public preview, giving AI agents tools to edit semantic models, generate DAX, validate queries, and answer questions from data. - Microsoft says the local Modeling MCP Server can create and manage tables, columns, measures, and relationships, while the hosted Remote MCP Server uses Copilot’s DAX engine and enforces user permissions. - The release folds Power BI into the Model Context Protocol push now spreading across developer tools, with preview caveats and warnings about inaccurate model changes. (learn.microsoft.com)
Power BI now has two Model Context Protocol servers in preview: one for editing semantic models and one for querying them with AI agents. (learn.microsoft.com) A semantic model is the layer that tells Power BI what a sales table, a margin measure, or a relationship between tables actually means. Microsoft’s new local Modeling MCP Server lets an AI agent change that layer with natural-language commands. (learn.microsoft.com) (github.com) Microsoft says the Modeling MCP Server can create, update, and manage tables, columns, measures, and relationships across Power BI Desktop and Fabric semantic models. It also supports bulk operations, best-practice checks, and DAX query validation. (learn.microsoft.com) (github.com) The second piece is the Remote MCP Server, a hosted endpoint for asking questions against an existing semantic model. Microsoft says AI assistants can pull the model schema, generate DAX from a natural-language prompt, and execute the query to return results. (learn.microsoft.com 1) (learn.microsoft.com 2) DAX, or Data Analysis Expressions, is Power BI’s formula language for measures and calculations. Microsoft says the Remote MCP Server uses the same DAX generation engine as Copilot for Power BI, and that tool requires a Copilot license unless teams disable it and rely on their own model. (learn.microsoft.com) The company tied the launch to its broader push for “context aware intelligence” in Power BI at Ignite in November 2025. In that announcement, Microsoft said Power BI serves more than 35 million monthly active users across 550,000 organizations. (powerbi.microsoft.com) The pitch is straightforward: let an agent inspect a model, write or test DAX, and make controlled changes instead of forcing analysts to click through every property by hand. Microsoft’s GitHub repository says those workflows can include TMDL and Power BI Project files for code-first model development. (github.com) (learn.microsoft.com) Microsoft’s own documentation also adds the caution. The MCP features are in preview, request formats and schemas can change, and the GitHub README warns that an underlying large language model can produce inaccurate results that lead to unintended model changes. (learn.microsoft.com 1) (learn.microsoft.com 2) (github.com) Security is built around the signed-in user rather than a separate AI permission layer. Microsoft says Remote MCP queries run with the authenticated user’s permissions, require at least Build permission on the semantic model, and enforce row-level security for user authentication. (learn.microsoft.com) That makes the story less about a single viral demo and more about Microsoft formalizing AI access to the Power BI model itself. The tools are real, they are public preview, and Microsoft is telling teams to review every machine-made change before they save it. (learn.microsoft.com) (github.com)
