Power BI Agentic AI: Claude & MCP Tips

Key insights

• Agentic AI demo: The video shows how to use Power BI MCP to connect Claude to an open Power BI Desktop model and let the agent actively modify the semantic model.
  It highlights practical prompts that save hours of repetitive modeling and documentation tasks.
  
• Connect & verify: First, connect Claude through the MCP server and confirm the connection by listing tables in the model.
  This step ensures the agent can read and act on the live Power BI file.
  
• Automated modeling: Use prompts to rename tables/columns to naming standards, hide key columns to reduce clutter, create sensible hierarchies, and generate organized measures and folders.
  The agent follows rules like “if unsure, don’t” to avoid unwanted changes.
  
• Relationships & DAX: The agent can create missing relationships (choosing the right date like order date) and build a composite key calculated column with DAX to link temperature and sales data automatically.
  This enables complex joins that would otherwise take manual DAX work.
  
• Data ingestion & self-correcting steps: You can add a new data source and let the agent import and self-correct steps when needed, keeping model structure clean and consistent.
  The agent also adds descriptions to improve usability for report authors.
  
• Power Query M & documentation: The workflow documents Power Query steps and adds Advanced Editor comments, then generates full documentation (markdown) that you can export to PDF.
  This centralizes model notes, relationships, measures, and data sources for handoffs and audits.
  

Video Overview and Context

In a new tutorial video from Pragmatic Works, the presenter returns for part two of a practical series that demonstrates how to use Agentic AI with Power BI, a MCP server and Claude Desktop to modify a semantic model. The walkthrough focuses on real prompts and live edits that automate repetitive modeling and documentation tasks, aiming to save hours for report authors and data modelers. Importantly, the video emphasizes hands-on examples rather than abstract theory, so viewers can follow along with a working Power BI Desktop file. As a result, the piece serves as a useful reference for teams exploring automation inside BI workflows.

Connecting Claude to Power BI via the MCP Server

First, the video shows how to connect Claude to an open Power BI Desktop file through the MCP server and then verify the connection by listing tables in the model. This step-by-step process clarifies how agents gain context about the semantic model before taking action, which reduces the risk of unintended changes. Moreover, the presenter highlights simple verification prompts to confirm connectivity and model visibility, which help teams avoid silent failures. Therefore, establishing a reliable connection upfront proves essential for safe automation.

However, while connection steps speed up repetitive actions, they raise tradeoffs around control and governance because automated agents can change production models quickly. Consequently, teams must balance convenience with safeguards such as version control, change approval workflows, and access limits. In addition, documenting the connection and any service accounts used helps maintain an audit trail for later review. Ultimately, adopting this approach requires investment in procedural checks as well as technical setup.

Modeling Automation Demonstrated

The core of the video demonstrates practical modeling edits driven by prompts: renaming tables and columns to match naming standards, hiding key columns to simplify the field list, and creating hierarchies where they are meaningful. The presenter also uses prompts to detect and create missing relationships, including selecting the correct Date ↔ Sales relationship by order date, which illustrates how agents can make contextual decisions. As a result, these tasks show clear time savings for routine cleanup and standardization across models.

Nevertheless, automating modeling introduces tradeoffs between speed and accuracy because an agent might apply a naming convention in a way that breaks downstream reports or hides columns that analysts still need. Therefore, the video recommends conservative prompts such as “if unsure, don’t” when creating structural changes like hierarchies. By contrast, manual review remains necessary for edge cases and business rules that require human judgment. Thus, combining agentic automation with reviewer checkpoints yields the best balance.

Advanced Scenarios: Composite Keys and DAX

In a more advanced segment, the presenter builds a composite key calculated column with DAX to relate regional temperature data to internet sales and then creates the relationship automatically. This example shows that agents can handle nontrivial transformations and generate DAX expressions to bridge different datasets, which can be especially useful in enrichment scenarios. Furthermore, the walkthrough adds a new table from an external source and demonstrates how the workflow can self-correct when initial prompts need refinement. Consequently, the video gives confidence that agents can scale beyond simple edits into complex modeling tasks.

Even so, complex solutions carry costs: composite keys and calculated columns may affect model performance, refresh times, and maintainability, so teams should weigh benefits against long-term overhead. In practice, developers must test performance impacts and document the logic clearly, because automated DAX generation can obscure intent. Therefore, governance and performance testing remain crucial when opting for automated advanced modeling steps. In short, power comes with responsibility.

Documentation and Export Workflow

Beyond modeling, the video illustrates how the MCP approach accelerates documentation by generating descriptions for tables, columns and measures and by adding annotations to Power Query M steps and the Advanced Editor. Additionally, the presenter compiles a full markdown documentation file that captures relationships, measures with DAX, data sources and other notes, and then exports the file to PDF for distribution. This capability streamlines knowledge transfer and ensures that changes made by agents remain visible to analysts and auditors. Therefore, documentation automation reduces the manual burden and supports traceability.

However, automated documentation also requires review to prevent inaccuracies and to capture business context that a model cannot infer. As a result, teams should use generated documentation as a draft that a subject matter expert verifies and enriches. Also, integrating these outputs into existing release processes improves governance and reduces the risk of mismatches between the live model and its documentation. Thus, the video frames documentation automation as a powerful aid that still depends on human validation.

Practical Takeaways and Challenges

Overall, Pragmatic Works offers a pragmatic, hands-on guide showing how Agentic AI with Power BI, MCP, and Claude can accelerate both modeling and documentation. The most tangible benefits include faster cleanups, repeatable edits, and quick generation of documentation that typically consumes much analyst time. At the same time, the tutorial makes clear that teams must manage tradeoffs between speed and control by introducing governance, testing, and review steps into automated workflows.

Finally, for organizations considering this approach, the message is straightforward: agentic tools can multiply productivity, but they do not replace human oversight. Therefore, blend automation with sensible limits, and monitor performance, permissions, and documentation to ensure that gains are real, repeatable, and safe. In this way, the video provides a balanced blueprint for getting started while highlighting the practical challenges that teams will need to address.

Keywords

Agentic AI Power BI, Power BI agentic automation, Claude AI Power BI integration, Power BI intelligent agents, Agentic analytics with Claude, Power BI MCP guide, Automating reports with Agentic AI, Claude-driven data workflows