Semantic-model work by conversation, including Desktop files.
Model work on your own machine is its territory, Power BI Desktop and project files included: refactors into calculation groups, naming clean-ups and translations. You can wrap a batch of edits in a transaction and roll it back, but Microsoft says to always back up the model before any operation.
What it is
The local version of Microsoft's Power BI Authoring MCP server, generally available. It runs on your own computer and lets an AI assistant build and change Power BI semantic models: tables, columns, measures, relationships and more. Unlike the hosted version, it also reaches models open in Power BI Desktop and Power BI Project files on disk. It changes models only, not report pages or diagram layouts.
What you get
- Connections to Power BI Desktop, a Fabric workspace or a Power BI Project folder
- Tables, columns, measures, relationships, partitions, hierarchies and calculation groups created, changed or deleted
- Security roles and row-level security, perspectives, cultures and translations managed
- DAX queries run and validated, and model activity traced
- Changes grouped in transactions you can commit or roll back
- Models imported and exported as TMDL folders, and deployed to Fabric
Requirements
- Node.js to run it with npx, or Microsoft's VS Code extension or standalone executable; it does not run on macOS, where the hosted version is the option.
- You sign in with Microsoft Entra ID, or use a service principal for automated pipelines.
- For models in a Fabric workspace, the capacity's XMLA endpoint must be set to Read Write, and you need Write permission on the model.
- Register either this or the hosted version, not both.
- Microsoft says to back up the model first; for a look-only session, start it in read-only mode.
Setup effort
One command plus credentials — npx -y @microsoft/powerbi-modeling-mcp@latest --start, then supply credentials