Agent Skills for Power BI: Complete Setup & Tutorial
Agent Skills for Power BI lets GitHub Copilot, VS Code Copilot, Claude Code, and Cursor build and refine your Power BI semantic models and reports using natural language. One plugin install. No manual MCP wiring. From a prompt to a deployed Direct Lake semantic model in minutes. This guide covers every step from install to production — verified against Microsoft Learn June 2026.
Agent Skills for Power BI (also called Power BI Agentic) is a curated bundle of agent skills and tools announced at Build 2026 that lets AI coding agents build and refine Power BI semantic models and reports using natural language. Install the powerbi-authoring plugin from the Skills for Fabric marketplace. It bundles five skills (semantic model authoring, report authoring, report design, report planner, report management) and automatically registers the Power BI Modeling MCP server. Works with GitHub Copilot CLI, VS Code Copilot, Claude Code, Cursor, Codex/Jules, and Windsurf. (per Microsoft Learn Power BI Agentic Overview, June 4, 2026)
What Agent Skills for Power BI Actually Is — and What It Is Not
Agent Skills for Power BI is not a button inside Power BI Desktop. It is not a Copilot feature you toggle on. It is a developer tool — a plugin you install into your AI coding agent so that agent gains the ability to understand Power BI’s authoring model, connect to your semantic models, and execute changes on your behalf.
Per Microsoft Learn (June 4, 2026): Power BI Agentic combines two building blocks:
- Agent skills — folders of instructions, scripts, and resources that an agent loads on demand. Skills tell the agent what to do: how to design a star schema, how to write DAX correctly, how to structure a PBIP project, how to validate a PBIR report definition. The agent reads the skill, loads its context, and applies it to your specific request.
- Tools — MCP servers and CLI bridges that let the agent actually do it: inspect schemas, run DAX queries, edit semantic models, validate reports, drive Power BI Desktop directly including reloading a model and capturing screenshots for visual verification.
The mental model: a skill is the playbook. A tool is the hand that executes the play. Both are required for the agent to author Power BI content reliably.
Why This Matters for Power BI Development Speed
The most time-consuming parts of semantic model development — designing the star schema, writing correct DAX measures, configuring relationships and cardinality, preparing the model for Copilot — are now prompt-driven. A developer who previously spent 3 hours manually building a Direct Lake model on a Fabric Lakehouse can do it with a single well-constructed prompt and a review cycle.
The November 2025 Power BI Modeling MCP release was the foundation — it gave AI agents the ability to interact with Power BI’s semantic layer programmatically. Agent Skills for Power BI (Build 2026) is the skills layer on top: the instructions that tell agents not just how to connect to a model, but how to author one correctly following Microsoft’s best practices.
Prerequisites — What You Need Before Installing
Per the official prerequisites from Microsoft Learn, two things are required before installing the powerbi-authoring plugin:
| Prerequisite | Version / Detail | Install Link |
|---|---|---|
| GitHub Copilot CLI | Required to run Skills for Fabric and the powerbi-authoring plugin from the command line | docs.github.com/copilot |
| Node.js | Version 18 or later — required to install the Power BI Modeling MCP server (registered automatically by the plugin) | nodejs.org |
| A semantic model (optional for install, required for authoring) | Power BI Desktop with a model open, a PBIP project on disk, or a semantic model in a Fabric workspace | Power BI Desktop or Fabric workspace |
If you prefer VS Code over the CLI, install the plugin through VS Code’s agent customizations — search for powerbi-authoring in the agent customization settings. The install scripts configure the appropriate compatibility files automatically for each supported agent.
Node.js 18+ Is a Hard Requirement
The Power BI Modeling MCP server is a Node.js package. If you install the powerbi-authoring plugin without Node.js 18 or later, the MCP server registration will fail silently and the semantic-model-authoring skill will fall back to slower TMDL file editing mode. Run node --version before installing to confirm your Node version.
Installing the powerbi-authoring Plugin — Step by Step
Method 1: GitHub Copilot CLI (Recommended)
- Add the Skills for Fabric marketplace This registers Microsoft’s skills marketplace as a source for copilot plugins. Run this once — it persists across sessions.
- Install the powerbi-authoring plugin This installs the full plugin — all five skills plus the Power BI Modeling MCP server. The fabric-collection tag ensures you get the curated bundle.
- Verify skills are loaded Start GitHub Copilot and run /skills. Look for semantic-model-authoring in the list. If it appears, your installation succeeded and the MCP server is registered.
- Confirm MCP server is running The MCP server should be registered automatically. Verify with copilot mcp show — powerbi-modeling-mcp should appear in the list and show as active.
Method 2: Visual Studio Code
In VS Code, open the Command Palette → Customize AI in Visual Studio Code → search for powerbi-authoring in the agent customizations editor. Select and install. The install scripts automatically configure VS Code Copilot compatibility shims.
Other Supported Agents
Skills for Fabric installs compatibility shims for Claude Code, Cursor, Codex/Jules, and Windsurf automatically as part of the plugin install. No separate configuration is needed — the install script detects your environment and sets up the appropriate compatibility files.
GitHub Copilot CLI
Primary target. Full feature support.
VS Code Copilot
Supported via agent customizations editor.
Claude Code
Compatibility shim installed automatically.
Cursor
Compatibility shim installed automatically.
Codex / Jules
Compatibility shim installed automatically.
Windsurf
Compatibility shim installed automatically.
All Five Skills — What Each One Does
Per Microsoft Learn, five skills are bundled in the powerbi-authoring plugin. The full and growing skill list lives at github.com/microsoft/skills-for-fabric.
Semantic Model Authoring
Create, edit, deploy, and manage Power BI semantic models across Power BI Desktop, PBIP projects, and the Fabric service. Covers Import, DirectQuery, and Direct Lake models. DAX measures, tables, columns, relationships. Deployment, refresh, data sources, parameters, and permissions. DAX performance optimization. AI readiness preparation for Fabric Copilot and Power BI Data Agents.
Report Authoring
Create, edit, and validate Power BI reports in PBIR (Power BI Report) format used by PBIP (Power BI Project) files. Build pages and visuals, apply formatting, validate report definitions. The agent reads and writes PBIR JSON directly — no Power BI Desktop UI interaction required.
Report Design
Produce a structured design brief for a Power BI report in PBIR format. Use before authoring to get a layout and visual recommendation from the agent before any code is written. The design brief guides the subsequent report-authoring workflow.
Report Planner
Guided workflow to define, plan, and build a new Power BI report from an existing semantic model. Starts with business questions, maps them to available measures and dimensions, then produces a report plan before executing the authoring step.
Report Management
Get, publish, and manage Power BI reports in Microsoft Fabric workspaces. List reports, publish new versions, manage report metadata. Useful in CI/CD pipelines where the agent manages report lifecycle automatically after content is authored.
Invoke Skills Explicitly with Slash Commands
To ensure a specific skill is loaded by the agent, invoke it with a slash command prefix: /semantic-model-authoring [your prompt] or /power-bi-report-authoring [your prompt]. Without the slash command, the agent may not select the right skill during progressive disclosure — particularly for ambiguous prompts. The slash command is the reliable way to load the exact skill you need.
Tools — The MCP Server and Desktop Bridge
Skills tell the agent what to do. Tools let the agent do it. Two tools are included in the powerbi-authoring plugin:
| Tool | What It Does | Auto-Registered? |
|---|---|---|
| Power BI Modeling MCP Server | Local MCP server that lets the agent inspect schemas, run DAX queries, edit semantic models in Power BI Desktop or Fabric workspaces, and manage data sources and permissions. Connects via the standard MCP protocol so any MCP-compatible agent can use it. | Yes — installed automatically by the powerbi-authoring plugin |
| Power BI Desktop Bridge | Local server hosted within the Power BI Desktop process. Enables external tools to communicate with a running Power BI Desktop instance — reload models, capture screenshots for visual verification, read the current model state. Required for authoring workflows that target a live Desktop session rather than a PBIP file or Fabric workspace. | Available separately — enabled when Power BI Desktop is open |
How the MCP Server Routing Works
Per Microsoft Learn, the semantic-model-authoring skill has a Tool Selection Priority that routes operations in this order:
- Power BI Modeling MCP server — if registered and reachable, all authoring operations go here first. Most reliable and full-featured path.
- Direct TMDL file editing — if MCP server is not available but a PBIP project exists on disk, the skill edits TMDL files directly.
- REST API round-trips — if targeting a Fabric workspace directly (getDefinition / updateDefinition). Slower than MCP but works without a local Desktop session.
- Not supported — authoring against Power BI Desktop with no PBIP project and no MCP server. The agent stops and asks you to register MCP or save the model as PBIP.
Best Practice — Always Use MCP Server + PBIP
The most reliable authoring setup: Power BI Desktop open with a PBIP project saved, MCP server running (registered by the plugin). This gives the agent the highest-fidelity connection — it can read the live model state, make edits through MCP, verify changes, and reload the model, all in one session. TMDL file editing works but requires a manual Desktop reload after each change.
Semantic Model Authoring — Core Workflows and Example Prompts
The semantic-model-authoring skill covers a broad set of semantic model tasks. Per Microsoft Learn, these are the primary workflows the skill supports:
Analyze Model Against Best Practices
The agent loads the skill, connects to Desktop through the MCP server, inventories the model, and reports findings grouped by severity. If this prompt works end to end, your entire setup is correctly configured.
Create DAX Measures
Configure Relationships
Optimize DAX Performance
Deploy to Fabric Workspace
Report Authoring — Build Power BI Reports with Natural Language
The report authoring skills work on PBIR (Power BI Report) format — the JSON-based report format used by PBIP (Power BI Project) files. The agent reads and writes PBIR files directly, without interacting with the Power BI Desktop UI.
Recommended Workflow: Plan → Design → Author → Publish
PBIR Is the Future — Enable It Before Authoring
The report authoring skill works on PBIR format exclusively. If your report is still in the legacy .pbix format, convert it to PBIP (Power BI Project) first: File → Save As → Power BI Project (.pbip). PBIP saves the report definition as PBIR JSON files that the agent can read and write. Per the PBIR guide, PBIR is moving toward becoming the only supported report format — migrating now positions your team for the agentic authoring workflow.
Complete Walkthrough — Build a Direct Lake Model on a Fabric Lakehouse
This is the complete end-to-end walkthrough from Microsoft Learn’s official example — verified against the June 5, 2026 documentation.
- Prepare the Fabric Lakehouse In the Fabric portal: create a workspace (e.g., SalesAnalytics). In that workspace, create a Lakehouse (e.g., SalesLakehouse). Load it with data — either select “Start with sample data” for built-in datasets (Public holidays, NYC Taxi) or upload your own files to the Files section and promote them to Tables.
- Start GitHub Copilot CLI in any folder You do not need to be in a Power BI Desktop session for this workflow. The agent connects to the Fabric workspace remotely via the MCP server’s REST API path.
- Run the creation prompt The skill runs the Create New Semantic Model workflow — resolves workspace and Lakehouse IDs, discovers the schema, creates a Direct Lake model following modeling guidelines, and deploys it to the workspace.
The agent executes these steps automatically:
- Resolves workspace and Lakehouse IDs via Fabric REST API
- Discovers the Lakehouse schema — tables, columns, data types
- Creates a Direct Lake semantic model following Microsoft modeling guidelines (star schema where possible, relationships from discovered foreign key patterns, appropriate data types)
- Deploys the model to the target workspace
- Verify in Fabric portal Open the SalesAnalytics workspace and confirm the new semantic model appears in the item list. Open it and verify the tables, relationships, and Direct Lake connection are configured correctly.
- Build reports on the model Open Power BI Desktop, connect to the new semantic model via Live Connection to the Fabric workspace, and begin building reports — or use the power-bi-report-planner skill to let the agent plan and build the first report automatically.
AI Readiness — Prepare Your Semantic Model for Copilot and Data Agents
A semantic model that was built for human report developers may not be optimized for AI consumption. Fabric Copilot and Power BI Data Agents need well-described measures, clear table and column naming, complete descriptions, and correct relationships to produce trustworthy answers from natural-language queries.
The semantic-model-authoring skill includes a Semantic Model AI Readiness workflow that evaluates your model against the AI readiness checklist and applies fixes automatically where possible.
The agent executes this workflow:
- Loads the
semantic-model-ai-readiness.mdreference from the Skills for Fabric repository - Inventories the model — all tables, columns, measures, relationships, hierarchies
- Evaluates every item against the AI readiness checklist (descriptions present, synonyms configured, sensitive columns hidden, star schema intact, Q&A synonyms active)
- Reports findings grouped by severity — agent-applicable fixes vs user-action-required items
- Applies approved fixes through the MCP server and saves the model
- Recommends testing representative natural-language prompts in Copilot or your Data Agent after changes
What AI Readiness Actually Checks
- Descriptions on all tables and measures — Copilot uses these to understand what each object means before answering queries
- Column descriptions on key fields — especially date columns, key measures, and dimension attributes
- Synonyms in Q&A — alternative names users might say for table and column names
- Hidden columns and internal measures — items not relevant to business users should be hidden so the AI doesn’t surface them in answers
- Correct cardinality on relationships — many-to-one from fact to dimension, not many-to-many unless intentional
- Mark as Date Table — date tables must be correctly marked for time intelligence to work in AI-generated DAX
Fabric Apps for Semantic Models — What It Is and How It Connects
Agent Skills for Power BI is the authoring tool. Fabric Apps for Semantic Models is the deployment output — announced alongside Agent Skills at Build 2026 as a separate but complementary capability.
Per the official Build 2026 announcement: Fabric Apps for Semantic Models enable AI agents to build and deploy Fabric-native web applications directly on top of semantic models. The same semantic model that powers Power BI reports can now serve as the data and logic foundation for a full-blown web app — without rebuilding the business and governance logic in a separate backend.
| Capability | Agent Skills for Power BI | Fabric Apps for Semantic Models |
|---|---|---|
| What it does | Builds and refines semantic models and Power BI reports | Builds and deploys Fabric-native web apps on top of semantic models |
| Who uses it | Data modelers, analytics engineers, BI developers | App developers, full-stack developers, business app builders |
| Agent involved | GitHub Copilot CLI, VS Code Copilot, Claude Code, Cursor | AI coding agent (any) prompted to build web app from semantic model spec |
| Output | Semantic model (.tmdl) + reports (.pbir) deployed to Fabric workspace | Fabric-native web application deployed on top of a semantic model |
| Preview status | Preview (June 2026) | Preview (June 2026, announced Build 2026) |
The workflow: use Agent Skills for Power BI to build and validate the semantic model → use Fabric Apps for Semantic Models to deploy a web application on that model with persona-specific views, custom calendars, and business logic. The model is the foundation. The app is the interface.
Troubleshooting — Common Issues and Fixes
| Problem | Cause | Fix |
|---|---|---|
| Skill isn’t being loaded by the agent | Plugin not installed or not picked up by the agent during progressive disclosure | Verify installation with /skills. Use explicit slash command — /semantic-model-authoring [prompt] — to force the correct skill to load. |
| MCP server not loading | GitHub Copilot CLI needs restart to pick up new MCP config after install | Restart GitHub Copilot CLI. Run copilot mcp show to confirm powerbi-modeling-mcp appears and is active. |
| Agent falls back to TMDL file editing instead of MCP | MCP server not reachable or not registered | Ensure Power BI Desktop is open with the model loaded. Restart CLI. Re-run copilot plugin install powerbi-authoring@fabric-collection if MCP still doesn’t appear. |
| Node.js version error on install | Node.js version older than 18 | Run node --version. If below v18, install Node.js 18 LTS from nodejs.org and retry. |
| Agent stops and asks for MCP or PBIP | Trying to author against Power BI Desktop with no PBIP and no MCP — unsupported path | Save the model as PBIP (File → Save As → Power BI Project) or register the MCP server before continuing. |
| Skill loads but agent produces incorrect DAX | Model lacks descriptions, synonyms, or AI readiness configuration | Run the AI Readiness workflow first: /semantic-model-authoring Connect to Power BI Desktop and prepare the semantic model for AI |
Frequently Asked Questions
copilot plugin marketplace add microsoft/skills-for-fabric — adds the Skills for Fabric marketplace; (2) copilot plugin install powerbi-authoring@fabric-collection — installs all five skills and registers the Power BI Modeling MCP server automatically. Verify with /skills in GitHub Copilot — semantic-model-authoring should appear in the list.copilot mcp show — powerbi-modeling-mcp should appear as active. Restart GitHub Copilot CLI after plugin install if it doesn’t appear.Official References and Related UIG Guides
Official Microsoft Documentation
Related UIG Data Lab Guides
⚠️ Accuracy Disclaimer
All skill names, install commands, workflow steps, and tool descriptions are verified against Microsoft Learn Power BI Agentic Overview (updated June 11, 2026) and Semantic Model Authoring Skill overview (updated June 11, 2026). Agent Skills for Power BI is in preview — install commands, skill names, and behaviors may change before general availability. Verify current documentation at Microsoft Learn before building production workflows. UIG Data Lab is an independent publication, not affiliated with or endorsed by Microsoft Corporation.