Power Automate: Build Flows with Claude

Key insights

• What it does: Claude Code, paired with the Microsoft Power Automate plugin and the FlowAgent MCP server, turns plain-English prompts into real cloud flows so you can build, edit, validate, and run flows from a terminal.
  It removes the need to open the web flow designer for many common tasks.
• Key prerequisites: Install Node.js, Visual Studio Code, sign in with the Azure CLI, and enable the Power Platform tools or CLI in your environment.
  Ensure you have Power Platform environment access and the required roles and connections before creating flows.
• Setup and run steps: Add the Power Platform skills marketplace plugin to Claude Code, install the power-automate plugin, then run the integrated /setup to authenticate and pick a target environment.
  Answer Claude’s clarifying questions (for example: recipient or turn-on option) so the agent can build the flow correctly.
• Demo example: The tutorial builds a SharePoint-triggered email flow (“new list item → send Outlook email”) end-to-end, with Claude Code discovering connectors, composing the flow JSON, validating it, and creating the flow in under two minutes.
  After generation, you review, enable, and test the flow in the Power Automate portal.
• Capabilities and controls: FlowAgent exposes actions to list environments and flows, read and edit JSON definitions, inspect run history and per-action inputs/outputs, and manage connections.
  Microsoft emphasizes review and validation before publishing to keep the workflow safer and auditable.
• Practical benefits and recommendation: This terminal-native workflow speeds development, lets teams stay in code-focused environments, and supports governance scenarios via solution tooling.
  For heavy MCP workflows, a paid Claude Pro plan is recommended to avoid limits during large or frequent operations.

Overview of the video

In a clear, hands-on tutorial, Dhruvin Shah [MVP] demonstrates how to build a cloud flow in Power Automate using only natural language and a terminal session. He connects Claude Code to Microsoft Power Automate with the official Power Automate plugin and the FlowAgent MCP server, then generates a working SharePoint-triggered email flow in under two minutes. Consequently, viewers see the flow created, validated, and tested without opening the browser-based flow designer, making the process feel immediate and streamlined.

Shah structures the demo to be reproducible: he lists prerequisites, walks through installation and setup steps, and finishes by testing the generated automation. He also explains how Claude Code asks clarifying questions so the agent can choose connectors and settings accurately. This approach emphasizes practical, repeatable steps rather than abstract claims.

Setup and step-by-step demo

The video begins with a checklist of tools and permissions: Node.js, Visual Studio Code, the Azure CLI signed into the tenant, and Power Platform access with a System Customizer role. Shah then installs the Power Platform skills marketplace and the power-automate plugin inside Claude Code, runs the plugin setup command, and selects the target environment so subsequent actions land in the right tenant context.

After the initial configuration, Shah writes a plain-English prompt to create a flow that sends an Outlook email when a new SharePoint list item appears. The FlowAgent MCP server translates that prompt into a full flow definition, validates connectors, and generates the JSON for the cloud flow. Shah reviews the generated definition in the terminal, turns the flow on in the portal, and then triggers a test to confirm the email is sent.

How the integration works

Under the hood, the plugin exposes operations that let the agent list environments and flows, read full flow JSON, edit and publish flows, and inspect run history and action inputs and outputs. Therefore, the agent participates in the full lifecycle of a flow: planning, validation, creation, testing, and debugging, rather than only suggesting code snippets. Shah demonstrates these lifecycle steps from the terminal, showing how generated definitions can be inspected before publishing for a safer rollout.

Importantly, the workflow relies on several moving parts working together: Claude Code to interpret natural language, the FlowAgent MCP server to convert prompts into flow operations, and Microsoft’s plugin to authenticate and act in the Power Platform environment. As a result, correct authentication via the Azure CLI and valid connector tokens for services like Outlook and SharePoint are essential for success.

Benefits and tradeoffs

The main benefit Shah highlights is speed: building routine automations from plain English reduces context switching and accelerates prototyping. Teams can stay in a terminal workflow and quickly create, test, or inspect flows, which may boost productivity for developers and administrators who prefer command-line tools. Moreover, because generated definitions can be reviewed and validated before publishing, the approach aims to keep some human oversight in the loop.

However, this speed comes with tradeoffs. Relying on an AI-driven path requires careful governance because generated flows can introduce unexpected logic or connector usage, and prompts may omit important constraints. Also, Shah recommends a paid Claude Pro plan for MCP-heavy workflows, which adds a cost consideration compared with manual flow design in the portal.

Challenges, risks and best practices

There are practical challenges to balance: security and permissions, connector authentication, and environment governance all demand attention before automating at scale. For instance, service account permissions and tenant-level policies can block agent actions, and reviewers must verify that generated expressions and actions follow organizational rules. Shah emphasizes the need to review the JSON definitions and test runs to ensure the flow behaves as expected.

Another challenge is non-deterministic AI behavior; the agent may choose different connector operations for similar prompts, so teams must design prompts and validation steps to minimize surprises. Consequently, combining this terminal-native approach with established change control, peer review, and solution packaging via the Power Platform CLI helps maintain traceability and governance. In short, the workflow improves speed but requires disciplined oversight to manage risk effectively.

Takeaways for practitioners

Dhruvin Shah [MVP] presents a practical, repeatable path to create, validate, and test Power Automate flows from a terminal using Claude Code and the FlowAgent MCP server. The demo shows that natural language can drive real automation, while also making clear that prerequisites, permissions, and careful review remain essential. For teams, the approach represents a useful tool in the automation toolbox when balanced with governance and testing.

Overall, the video is a useful primer for developers and administrators who want to experiment with AI-assisted Power Platform workflows, and it highlights both the promise and the responsibilities that come with agent-driven automation. Readers should view this as a practical demonstration rather than a turn-key solution, and plan to combine rapid creation with deliberate validation and change control.

Keywords

Power Automate flows, Claude Code automation, FlowAgent MCP setup 2026, Build Power Automate flows, Claude integration Power Automate, Power Automate tutorial 2026, FlowAgent setup guide, Automate workflows with Claude