Build an MCP Server - Step by Step - in Microsoft Agent Framework

Description

In this tutorial, we dive into Model Context Protocol (MCP) and show how to expose AI agents built with Microsoft Agent Framework as MCP servers, then consume them directly inside Microsoft Foundry. We start with a simple agent, convert it into an MCP server, and test it using the MCP Inspector. From there, we level up by running MCP over Server-Sent Events (SSE), exposing it via HTTP, and making it accessible through Ngrok. Next, we tackle a common question: 👉 Can workflows from Microsoft Agent Framework be reused in Microsoft Foundry? We clarify the differences between workflows in each framework and demonstrate a practical solution—wrapping a workflow as a tool and exposing it through MCP. By the end of this video, you’ll understand: What MCP is and why it matters How to expose agents and workflows as MCP servers How to connect local MCP servers to Microsoft Foundry How Agent Framework and Foundry workflows differ (and how to bridge them) 📦 All code shown in this video is available in the GitHub repo linked below. https://github.com/kirkmcpherson/microsoft-agent-framework-demo If you find this helpful, don’t forget to like, subscribe, and check out the other Agent Framework and Foundry tutorials on the channel. 00:00 – Intro & Goal of the Tutorial 01:47 – Agents vs Workflows in Microsoft Agent Framework 02:21 – What Is MCP? (Model Context Protocol Explained) 03:05 – Converting an Agent into an MCP Server 06:19 – Testing MCP with the MCP Inspector 10:17 – Running MCP over SSE and HTTP 13:39 – Exposing MCP Locally with Ngrok 15:50 – Using Agent Framework Workflows 17:16 – Workflow Differences: Agent Framework vs Foundry 21:05 – Wrapping a Workflow as an MCP Tool 26:16 – Connecting MCP to Microsoft Foundry 30:03 – Wrap-Up & Next Steps #MicrosoftAgentFramework #ModelContextProtocol #MCP #MicrosoftFoundry #AzureAI #AIAgents #AgenticAI #MultiAgentSystems #ServerSentEvents #Ngrok #AzureOpenAI #AIWorkflows #Copilot #LLMOps