From Adaptive Card to AI: Connecting Copilot Studio Forms to Azure AI Foundry скачать в хорошем качестве

From Adaptive Card to AI: Connecting Copilot Studio Forms to Azure AI Foundry

This video tutorial is a continuation of a series on building clients that integrate with an Azure AI Foundry Agent and an MCP (Model Context Protocol) Server. The ultimate goal is to have multiple client interfaces that can submit data (like insurance policy or claim information), which is then processed by the AI agent, sent to the MCP server, and stored in Azure Blob Storage as both a generated video (MP4) and a metadata file (JSON). Core Architecture: The system is built on a multi-tier integration: Destination: Azure Storage Account (holds final JSON and MP4 files). MCP Server: The intermediary that interacts directly with the storage account. It's tested using Postman. Azure AI Foundry: Hosts the core "Insurance Policy Agent" (or similar) that contains the business logic and calls the MCP server. Clients (Multiple Front-ends): Client 1: A custom ASP.NET web application ("Insurance Claim Portal") that calls the Foundry Agent directly. Client 2: A Microsoft Form, which triggers a Power Automate flow, which then calls an Azure Function that interacts with the Foundry Agent. Client 3 (This Video's Focus): A Copilot Studio Agent, which collects user input via an Adaptive Card, triggers a Power Automate flow, which calls the same Azure Function to route data to the Foundry Agent and MCP Server. This Video's Objective: To build the third client—a conversational AI agent in Microsoft Copilot Studio—that can guide a user through submitting an insurance policy via a form, and seamlessly trigger the backend process to generate and store the output. Key Workflows Demonstrated: Designing an Adaptive Card form within a Copilot Studio topic to collect user input. Creating a Power Automate flow triggered by the Copilot agent to act as a bridge. Configuring the flow to call an existing Azure Function (HTTP trigger) with dynamic data from the form. The Azure Function authenticates and passes the payload to the Azure AI Foundry Agent. The Foundry Agent executes its logic (likely involving AI-based video generation from text) and calls the MCP Server. The MCP Server finally creates two files in Azure Blob Storage: a .json file with the input metadata and an .mp4 video file with synthesized speech based on the provided content. ⏱️ Timestamps: 0:00 - Introduction & Architecture Review Recap of previous clients (web app, Microsoft Forms) and today's goal: building a Copilot Studio client. 2:30 - Creating the Copilot Studio Agent & Topic Setting up a new agent and defining a topic for insurance policy submissions. 7:00 - Designing the Adaptive Card Form Using AI to generate and implement a JSON-based form to collect user input. 12:00 - Building the Power Automate Flow Creating a flow triggered by Copilot, configuring it to call the Azure Function with Key Vault authentication. 18:00 - Integrating Flow with Copilot Studio Connecting the flow to the topic, mapping input parameters, and testing the chat interface. 22:00 - Dynamic Payload & Final Testing Replacing hardcoded data with dynamic values from the form and verifying file generation in Azure Storage. 26:00 - Conclusion & Publishing Wrapping up and briefly showing how to publish the agent to Microsoft Teams.