
Microsoft MVP (Business Application & Data Platform) | Microsoft Certified Trainer (MCT) | Microsoft SharePoint & Power Platform Practice Lead | Power BI Specialist | Blogger | YouTuber | Trainer
This article summarizes a recent YouTube video by Dhruvin Shah [MVP] that demonstrates how an AI can build a real Power Apps canvas app end to end. The video shows a working Expense Tracker that opens and remains editable inside Power Apps Studio, and it starts from a blank machine so viewers do not need prior setup. Importantly, the author walks through every installation and configuration step, which makes the workflow repeatable for IT pros and makers. As a newsroom summary, we present the sequence, technical details, and practical tradeoffs without claiming authorship of the original content.
Dhruvin Shah begins by explaining how the Canvas Apps MCP Server connects an AI coding assistant, namely Claude Code, directly to the Power Apps authoring environment. He then builds an Expense Tracker by generating the app source in .pa.yaml format, validating it, and syncing it live into a Studio session while showing screens getting created. Consequently, viewers see the full flow from initial prompt to a working app that writes back to a SharePoint list. Moreover, Shah records the commands and the coauthoring steps so others can repeat the process precisely.
Furthermore, the video covers the developer-facing mechanics such as token usage, wait times, and background validation that happen during the build. Shah also uses the slash commands provided by the toolset, for example /canvas-app, to force a sync back into Studio and to drive iterative changes. He demonstrates adding an expense live and then verifying the entry in the SharePoint data source to prove end-to-end correctness. Finally, the author highlights the need for a paid AI plan when using Claude Code, which affects cost and access.
Unlike many tutorials, the video starts from installing the .NET 10 SDK on Windows and then adds Visual Studio Code with the Claude Code extension. Next, Shah installs the Power Platform Skills plugin marketplace and the canvas-apps plugin inside the extension, and he creates a blank canvas app in Power Apps Studio. He then creates a SharePoint list to act as the Expense Tracker data source and connects it to the app so data can be written back. Importantly, Shah shows how to enable coauthoring under Settings → Updates, which is mandatory for the MCP-based workflow to operate.
After that, he copies the Studio Designer URL and pastes it into the AI assistant so the MCP layer can extract environment IDs, app IDs, and cluster info. This automatic discovery reduces manual errors but also requires careful handling of URLs and permissions. Shah walks through creating the project folder in VS Code and then demonstrates the first natural language prompt that produces a multi-screen app plan. He then shows how to review the generated plan and to start the coauthoring session using the provided URL.
The video summarizes the four-step flow behind the scenes as Describe → Discover → Generate → Validate & Sync, which helps viewers understand the sequence of operations. In the Describe step, the assistant interprets user intent via natural language; next, Discover pulls metadata from the Designer URL; then Generate creates the .pa.yaml app source; finally, Validate & Sync checks constraints and pushes the code into Studio. This pattern keeps Power Apps Studio as the single source of truth while allowing prompt-driven creation to proceed.
Shah also explains that the MCP Server handles token exchanges and monitors background tasks while the assistant writes screens and controls. During generation, the tool reports progress, and makers can use slash commands to interact while the build runs. Therefore, the workflow mixes automated generation with human oversight to catch layout or logic issues early. The video demonstrates how validation catches many structural problems, but some UI tweaks still need follow-up prompts.
Using AI to generate a working canvas app speeds development, however, it reduces immediate hands-on control over small layout decisions and formula choices. On the one hand, makers gain rapid prototyping and consistent scaffolding; on the other hand, they must accept iterative correction cycles to tune forms and alignment. Additionally, relying on an AI service requires a paid plan and raises questions about governance, tokens, and access control especially when the tool reads environment URLs and app metadata. Therefore, organizations must weigh productivity gains against licensing costs and data governance rules.
Moreover, keeping Power Apps Studio as the source of truth means the platform still enforces validation, which reduces the risk of pushing invalid app definitions but can cause delays when complex custom logic is needed. Likewise, teams should consider version control and coauthoring conflicts because multiple agents or users editing simultaneously can create merge challenges. Consequently, the approach works best for straightforward apps and for teams willing to combine AI output with manual refinements. In short, the method promises speed but requires policies and skilled oversight.
Shah highlights real obstacles such as layout issues, form control alignment, and formula complexities that the AI may not handle perfectly on the first pass. He shows how iterative prompts fix many issues and how to run the /canvas-app command to force a sync into Studio when needed, which underlines the interactive nature of the process. Patience matters because token usage and background processing introduce wait times, and sometimes the assistant needs multiple cycles to resolve validation errors. Thus, practical success depends on a mix of good prompts, testing, and manual adjustments.
Finally, Shah’s live demo of the final Expense Tracker proves the concept by adding an expense and verifying it in SharePoint, which reinforces the claim that AI can produce an editable canvas app, not just a mockup. For readers hoping to try this, the main takeaways are to prepare the prerequisites, enable coauthoring, and expect iterative fixes rather than perfect first-pass results. Overall, the video offers a clear, repeatable path for makers and IT teams who want to explore prompt-driven Power Apps development while understanding the tradeoffs.
Power Apps Canvas App tutorial, Claude Code Power Apps integration, MCP server setup 2026, Build Canvas App 2026, Power Apps with Claude AI, Power Platform Claude code guide, MCP server configuration for Power Apps, Low-code Power Apps step-by-step