Agent Skills for Power BI: Current Setup & Tutorial

What Power BI Agentic Does — and What It Doesn’t

Power BI Agentic is not a single toggle in Power BI Desktop. It is a development workflow assembled from skills, agent tooling, and Power BI authoring interfaces. You describe a task in natural language; the agent uses the relevant skill to plan the work and uses tools to inspect or change the model or report.

A skill is guidance, not a guarantee of correctness. Model relationships, generated DAX, visual bindings, permissions, and deployment results still need engineering review and validation.

The Current Skill Structure

The current Microsoft Learn overview lists two Power BI skills in the powerbi-authoring bundle. The September 17, 2026 Skills for Fabric release replaced four separate report skills with the unified powerbi-report-cli skill. The report workflow still has planning, design, authoring, and management capabilities, but they are modules routed through one skill—not four separate skills to install.

Report capabilities inside powerbi-report-cli

Prerequisites and Installation

Microsoft’s recommended quick start uses GitHub Copilot CLI and the Skills for Fabric marketplace. The setup scripts also provide compatibility configuration for other supported agents. The exact UI and command experience can vary by agent version.

Method 1 — GitHub Copilot CLI (recommended quick start)

1. Add Microsoft’s Skills for Fabric marketplaceRun this in a terminal with the current GitHub Copilot CLI installed.

1. Install the Power BI authoring pluginThis installs the current skill bundle and registers the local Power BI Authoring MCP server.

1. Open Copilot and verify the skillsRun the skills command in the Copilot session and confirm both current Power BI skills are available.

1. Check the MCP connectionWithin the Copilot CLI session, inspect MCP status. If the server is missing or disconnected, restart the CLI and follow the current MCP troubleshooting documentation.

Method 2 — Visual Studio Code

In VS Code, open Customize AI in Visual Studio Code from the Command Palette or the agent customizations experience, then search for powerbi-authoring and follow the installation flow described in Microsoft’s Power BI Agentic overview.

Supported agent environments

Microsoft optimizes Skills for Fabric for GitHub Copilot CLI. Compatibility shims do not imply every agent/version has identical UI, authentication, approval behavior, or feature parity.

Choose the Right Power BI Authoring MCP Server

The Power BI Authoring MCP server provides tools for creating and changing semantic models. Microsoft documents two deployment options for this same authoring server: a Microsoft-hosted endpoint and a local server. They have different access boundaries.

Manual hosted server configuration

Use this only when intentionally configuring the hosted endpoint. The tenant setting Users can use the Power BI Model Context Protocol server endpoint must be enabled by a Fabric administrator, and you must have the appropriate semantic model permissions. Do not add the hosted server if the current plugin configuration already supplies the local server for this session.

Manual local server configuration

The current Microsoft Learn guide documents an npm-based local setup. For plugin users, the local MCP server is normally configured by the plugin; use manual registration only when you need a separate setup or are troubleshooting.

Before asking the agent to modify a model, explicitly identify the target. For a Fabric model, name both the semantic model and workspace. For Desktop, identify the open model. For a PBIP project, provide the path to the model definition folder. Then begin with a read-only check such as “List the tables and measures in this model.”

The Power BI Desktop Bridge is another local component. It runs within Power BI Desktop and allows compatible tools to interact with a running Desktop instance, including reload and screenshot-based verification. It is distinct from the semantic-model Authoring MCP server.

Semantic Model Authoring: Reliable Prompts That Include Validation

The best prompts name the target model, specify the requested change, ask the agent to inspect existing metadata first, and require verification. Avoid asking an agent to invent table names or relationships based on guesses.

1. Run a setup smoke test

This checks whether the skill can load, the target model can be identified, and the tools are reachable. If using a Fabric workspace or PBIP project instead, name that target explicitly.

2. Create a DAX measure without guessing the schema

This is safer than naming a hypothetical Sales table or assuming that a date table is correctly marked. The agent should use the real model metadata and state any prerequisites.

3. Review relationship design

4. Analyze DAX performance

5. Document the model

Power BI Report Authoring with the Unified Skill

The current powerbi-report-cli skill covers report planning, design, authoring, validation, preview, and management. It works with local PBIP projects using PBIR report definitions and with supported report items hosted in Microsoft Fabric. Local file-based authoring requires a PBIP project using PBIR; this does not mean every report-management task requires converting an existing report first.

Recommended workflow: Plan → Design → Author → Validate → Manage

1. Plan the reportIdentify audience, business questions, KPIs, page scope, model dependencies, and delivery target. Review the proposed specification and approval checkpoint.
2. Design the pagesDefine each page’s purpose, hierarchy, chart choices, filters, interactions, theme, typography, and accessibility requirements.
3. Author the PBIR definitionCreate or modify pages, visuals, slicers, filters, bookmarks, bindings, and formatting using the actual semantic model fields.
4. Validate both structure and rendered outputRun validation, preview the report, inspect screenshots, and check for empty visuals, errors, broken bindings, or accessibility issues. Structural validation alone is not visual validation.
5. Publish only after reviewConfirm the destination workspace, report/semantic-model binding, and deployment effects. Check the final item in Fabric.

Local report-authoring requirements

• Use a PBIP project with PBIR report definitions for local file-based authoring.
• Commit a baseline or create a backup before substantial agent edits.
• Save manual changes made in Power BI Desktop before asking the agent to modify the project; local files are the source of truth for local authoring.
• Use rendered-output verification after meaningful changes. Microsoft’s current guidance describes a Power BI Desktop preview when available and a headless-browser path otherwise.
• Prefer supported modern visuals. Avoid visuals Microsoft documents as scheduled for deprecation, including Q&A, Bing Maps, and filled maps.

Walkthrough: Create a Direct Lake Semantic Model on a Fabric Lakehouse

This workflow follows Microsoft’s semantic-model-authoring example. A real result depends on workspace access, available Fabric capacity and features, source data, tenant configuration, and semantic-model permissions; a prompt alone does not bypass those requirements.

1. Prepare a Fabric workspace and LakehouseCreate or select a workspace, create a Lakehouse, and load usable data into its Tables area. Sample data can be used to test the workflow.
2. Confirm access and target namesEnsure you can access the workspace and create the required semantic model. Record the exact workspace and Lakehouse names and verify that the tables contain the columns you expect.
3. Ask the semantic model skill to create the modelUse the official example pattern. The agent should resolve the workspace and Lakehouse, inspect the schema, create the model, and deploy it if the available tools, permissions, and capacity support those actions.

Verify the result instead of assuming success

• Confirm that the semantic model appears in the intended Fabric workspace.
• Inspect tables, columns, data types, relationships, and storage mode.
• Validate important measures and totals against source data or known results.
• Review refresh/processing status, permissions, and any Direct Lake restrictions or fallback behavior relevant to the model.
• Only then build a report using powerbi-report-cli or Power BI Desktop.

Prepare a Semantic Model for AI Consumption

A model intended for human-authored reports may need clearer metadata and tighter modeling conventions before users query it through natural language. The semantic-model-authoring skill includes an AI-readiness workflow that assesses the model and proposes changes for review.

What to review

• Business descriptions: tables, important columns, and measures need definitions that reflect the organization’s actual business terms.
• Measure semantics: validate formulas, filters, time intelligence, and formatting; don’t infer business definitions from object names alone.
• Relationships: verify cardinality, filter direction, uniqueness assumptions, and ambiguous paths.
• Hidden technical objects: hide fields users should not select where appropriate, while preserving legitimate technical dependencies.
• Date modeling: confirm the date table, date-column mapping, and time-intelligence prerequisites are set correctly.
• Privacy and access: ensure sensitive columns, row-level security, object-level security, and user permissions are appropriate for the target AI experience.

Run representative test questions after approved changes and compare answers with validated DAX or report results. AI readiness helps the agent interpret a model; it does not guarantee the accuracy of all natural-language answers.

Security, Permissions, and Production Guardrails

These tools can modify semantic models and reports. Treat agent output as a proposed change set, not as automatically trusted production code.

Permissions that matter

For the Authoring MCP server, Microsoft’s documentation says Write permission is needed to create or change model objects; Build-only access is limited to querying the model with DAX. The hosted server additionally requires the tenant setting that allows use of the Power BI MCP endpoint. Local access to workspace models can require XMLA endpoint configuration, depending on the deployment and target. Confirm the current prerequisites in the official Authoring MCP setup guide.

Fabric Apps are a separate capability

Fabric Apps for semantic models are a separate preview capability for creating custom web applications on top of semantic models. Agent Skills for Power BI focus on the semantic model and report development workflow; Fabric Apps focus on custom application experiences. One is not simply the deployment output of the other.

Troubleshooting: Common Issues

Frequently Asked Questions

Official Microsoft References

These primary sources were checked for this revision. Microsoft Learn and the Skills for Fabric repository can change; verify commands and current limitations before using them in production workflows.

Related UIG guides