TMDL

How To Use Claude Code for Power BI Development

How do you use Claude Code for Power BI development? Open your Power BI project (PBIP) folder in VS Code, run Claude Code in the built-in terminal, and describe the change in plain English: the agent reads the project files, writes the DAX, edits the Sales.tmdl file itself, and asks your permission before each command. Codex from OpenAI works the same way; this is a walkthrough for anyone who has only ever copy-pasted DAX out of a chat window.

Related videos

What you'll learn

  • Open the terminal inside VS Code (Ctrl+J on Windows, Cmd+J on Mac) and run basic commands

  • Start Claude Code with the claude command and hand it a job in plain English, such as adding a new measure to a table

  • Answer the permission prompt (yes, yes and don't ask again, or no) and switch modes with Shift+Tab: manual, accept edits, plan, auto, and bypassPermissions

  • Point the agent at one file with an @ tag so it edits exactly the TMDL file you mean

  • Call a reusable skill with /, tell local skills from global ones, and vet skills from third parties before you install them

Covered

Terminal basics, VS Code integrated terminal, AI harnesses, Claude Code, Codex, permission prompts and modes, Power BI project (PBIP) folders, TMDL model files, DAX measures, the context window, @ file tags, skills, Skills for Fabric

Tools

  • Claude Code (Anthropic), run from the VS Code terminal

  • Visual Studio Code

  • Power BI Desktop, with the report saved as a Power BI project (PBIP)

  • PowerShell on Windows, or Terminal on Mac

  • Codex (OpenAI), the equivalent harness if that is what your company allows

Prerequisites

  • A Power BI report saved as a Power BI project (PBIP), so the semantic model and report exist on disk as text files

  • VS Code installed, plus a Claude or ChatGPT account to sign in to the harness

  • Enough DAX knowledge to read a measure the agent writes and judge whether it is right; no terminal or coding experience needed

Resources

Subscribe to The Data Practitioner on YouTube

How to Fix Your Live Power BI Model Straight from the Browser

How do you fix a live Power BI semantic model without opening Desktop? Open it in the Service and use TMDL View on the web (Preview, July 2026 update): script the change, preview the diff, switch to Edit mode, and apply. The fix lands on the deployed model in seconds, with no republish and no data reload.

Related videos

What you'll learn

  • Hotfix a measure on a published semantic model straight from the browser: script it, preview a side-by-side diff, and apply in Edit mode

  • Add row-level security roles by pasting a TMDL script and validating them in the service

  • Bulk rename across your entire semantic model with find and replace

  • Tell which changes land instantly (metadata only) from the ones that wait on a data refresh

  • Restore a previous version with Version History, a rollback safety net Power BI Desktop does not have

Covered

TMDL View on the web, the Power BI Service, semantic model hotfixes, metadata-only apply, row-level security, bulk rename, version history, ephemeral script tabs, web-first development

Tools

  • Power BI Service, TMDL View (Preview, July 2026 update)

  • Any modern browser, including on a Mac (no Power BI Desktop required)

Prerequisites

  • Comfort reading and editing TMDL; this is not a TMDL introduction

  • Write permission on a published semantic model to follow along

  • Working knowledge of measures and row-level security in Power BI

Resources

Subscribe to The Data Practitioner on YouTube

Easy Version History Comes to Power BI Desktop

Power BI Desktop's May 2026 update adds built-in version history as a preview feature: save your .pbix to OneDrive or SharePoint and Desktop keeps a snapshot every time you save, so you can browse past versions, compare them side by side, and restore an older one without learning Git.

Related videos

What you'll learn

  • Turn on the Save to OneDrive and SharePoint preview feature and save a .pbix where version history works

  • Open the version history list from the file-name dropdown and see who changed each version, when, and the file size

  • Open a prior version next to the current one to spot what changed between them

  • Restore an older version with a single save, keeping every earlier snapshot intact

  • Recognize where file snapshots stop and real source control with Git and PBIP begins

Covered

Power BI Desktop version history, OneDrive and SharePoint saves, preview features, version snapshots, rolling back a report, comparing report versions, file versioning vs source control, solo and small-team workflows

Tools

  • Power BI Desktop, May 2026 update or later (version history is in preview)

  • OneDrive or SharePoint (where the version snapshots live)

Prerequisites

  • Comfortable building and saving .pbix files in Power BI Desktop

  • A OneDrive or SharePoint location you can save files to

  • No Git or source control experience needed

Resources

Subscribe to The Data Practitioner on YouTube

AI Can Now Check Its Edits in Power BI

How does your AI agent check the Power BI report it just edited? Point it at the Power BI Desktop Bridge, a local server that ships on by default in the June 2026 release of Power BI Desktop, so your agent can reload its PBIR edits, check for unsaved changes, and screenshot the report to grade its own work before you ever look.

What you'll learn

  • Connect an AI agent to Power BI Desktop through the Desktop Bridge and reload its PBIR edits without closing and reopening Desktop

  • Have the agent run the manifest command first so it only calls the functions your Desktop version actually supports

  • Use status to avoid overwriting unsaved work, and screenshot so the agent can visually verify its own edits

  • Drive an edit, screenshot, and grade loop from Claude Code to add conditional formatting to a matrix, then read the agent's own iteration log

  • Write looping prompts with specific, measurable acceptance criteria the agent can grade a screenshot against

Resources

Subscribe to The Data Practitioner on YouTube

How To Use Microsoft's Fabric Skills To Build a Power BI Report

Microsoft's official Skills for Fabric can author an entire Power BI report from one plain-English ask, the semantic model, the DAX measures, the page layout, and the publish to Fabric, and you run it yourself by pointing Claude Code at the open-source repo. This is the honest, unedited test: I hand the powerbi-authoring skills a vague "executive sales report" ask and let them run, stopping only where it teaches you something, including where the AI catches its own measure bug and the asterisk that a published report that can't refresh still isn't done.

What you'll learn

  • Point Claude Code at the open-source skills-for-fabric repo and run the powerbi-authoring plugin yourself

  • Drive a full report build from a single vague ask, with the planning skill interviewing you on scope and delivery before it writes anything

  • Read the Design Brief the AI commits to in plain YAML and approve it before a single visual exists

  • Let semantic-model-authoring write your TMDL tables, relationships, and measures into a PBIP, then review them like a fast junior analyst's work

  • Watch the report-authoring skill lay out the pages in PBIR, screenshot its own work, catch a year-over-year measure bug, and fix it on its own

  • Understand why a report published through the management skill lands empty, and what refresh and gateway work is still on you

Resources

Subscribe to The Data Practitioner on YouTube

Microsoft Shipped Open-Source AI Skills for Fabric

At Build 2026, Microsoft shipped an official, open-source repo of AI skills for Fabric that you install and run in your own AI tools (Claude Code, Copilot CLI, Codex, Cursor, Windsurf). They're written in prose, not code, so even a non-coder can read what they do, and in this first look one of them inventories and documents an entire Fabric workspace from a single one-line prompt.

What you'll learn

  • What microsoft/skills-for-fabric actually is, and why it signals Fabric going agentic in the open

  • How one one-line prompt ("Document the Samples workspace") discovers items, reads schemas, and writes the docs

  • The mental model that ties it together: connections, skills, personas, and bundles

  • How to install a bundle in Claude Code and wire up the Fabric MCP, including filtering to a single workload

Resources

Subscribe to The Data Practitioner on YouTube

The New Tabular Editor CLI That AI Agents Can Run

The new Tabular Editor CLI lets an AI agent operate your Power BI semantic model from the terminal, initialize a project, write C# scripts to add measures, run the Best Practice Analyzer to validate and auto-fix violations, deploy to the Service, and more, with every change landing as a reviewable git diff.

What you'll learn

  • Initialize a fresh TMDL model (or a full PBIP project) from nothing with a single te init command

  • Have an AI agent write and run a C# script to add a measure to an existing model

  • Run the Best Practice Analyzer headlessly, read the JSON findings, and auto-fix the fixable violations

  • Review every agent change as a line-level TMDL git diff before you keep it

  • Deploy the model to a Power BI workspace from the command line with te deploy and te refresh

Resources

Subscribe to The Data Practitioner on YouTube

The Power BI Modeling MCP Server — Hands-On

How do you let an AI agent actually edit your Power BI semantic model — not just hand you DAX to paste? You connect a Copilot- or Claude-style agent to your live model through Microsoft's official Power BI Modeling MCP server and the XMLA endpoint, then drive bulk model work (descriptions, hides, sort-bys, mark-as-date, new measures, even SVG-via-DAX visuals) in a fraction of the time — with the changes landing as a reviewable git diff when the model is saved as a Power BI project.

What you'll learn

  • Connect an AI agent to your live Power BI semantic model via the Modeling MCP server and the XMLA endpoint

  • Add descriptions to every table, column, and measure in your model in one prompt

  • Run an optimization pass where the agent proposes a plan first, you approve it, and then it executes

  • Verify what the agent actually changed by reading the TMDL git diff in a Power BI project

  • Roll back any change you do not like, file by file, using source control as your undo button

  • Generate an SVG sparkline visual by asking the agent to write the DAX measure for it

Resources

Subscribe to The Data Practitioner on YouTube

Get Your Power BI Reports AI-Ready with GitHub

A beginner-friendly setup walkthrough for getting a Power BI project into a GitHub repository so you can safely bring AI into your reporting workflow, with rollback and visible diffs as the safety net.

What you'll learn

  • Install Git and Visual Studio Code and configure Git with your name and email

  • Create a GitHub repository and clone it locally with VS Code

  • Save an existing Power BI report as a Power BI project (PBIP) inside the repo

  • Use a .gitignore to keep the local data cache out of the repository

  • Run a full lifecycle pass: branch, add a DAX measure, view the diff, commit, sync, and merge a pull request

Resources

Subscribe to The Data Practitioner on YouTube

Power BI as Code: The Setup AI Needs for Your Reports

Save your Power BI report as a PBIP project and it lands on disk as a folder of text — JSON for the report layer, TMDL for the model — which is the format AI tools like Claude and Copilot can actually read, reason about, and edit. This video walks through the three preview features that make that possible, then tours what each piece of the project looks like on disk.

What you'll learn

  • Enable the three Power BI Desktop preview features that make PBIP usable for AI workflows (PBIP save option, TMDL semantic model, enhanced report metadata)

  • Save an existing .pbix as a PBIP project and see what changed on disk

  • Read the .Report folder — how PBIR represents your report as files and folders of JSON

  • Read the .SemanticModel folder — how TMDL captures your data model as plain text

  • Use TMDL triple-slash (///) comments as model descriptions that round-trip from AI edits back into Desktop

Resources

Subscribe to The Data Practitioner on YouTube