Azure Logic Apps + MCP Servers: Full Build & Optimization Demo

Unlock the full power of your AI agents by moving tool logic out of Logic App Agent loops and into reusable Model Context Protocol (MCP) Servers. In this video, I demonstrate a new capability in Azure Logic App Agent Loops that lets you call both built-in Microsoft MCP servers and your own custom MCP servers. Previously, you had to define tools and actions directly inside the agent loop. Now, you can abstract that logic into standalone Logic Apps exposed as MCP servers, enabling better reusability across your enterprise. I walk through a complete demo of updating a "Five Card Draw" poker agent. You'll see how I replace manual tools with a custom MCP server to handle game logic (like "Ask for Help" and "Check Winner") and integrate the built-in Microsoft Email MCP server to send results. I also cover critical optimization tips, such as using the "Allowed Tools" property to limit context and improve performance. In this video, we cover: ✅ The Shift: Moving from internal Agent Tools to external MCP Servers. ✅ Custom MCP Servers: How to expose a Logic App as an MCP server. ✅ Integration: Adding a custom MCP connection to an Agent Loop. ✅ Microsoft Built-in Tools: Using the Email Management MCP server. ✅ Optimization: Configuring "Allowed Tools" to save tokens and cost. ✅ Live Demo: Running a Poker Agent with fully abstracted tool logic. ✅ Debugging: Viewing MCP server calls in the Logic App Run History. Reference from the product group: - https://www.youtube.com/watch?v=CnI_1E_rccs - https://www.youtube.com/watch?v=28sZX60N_fU 00:00 Intro to MCP Servers in Logic App Agent Loops 01:23 Overview of Five Card Draw Poker Agent 03:19 Creating Custom Logic App MCP Servers 05:00 Adding MCP Servers to Agent Loops 06:55 Configuring Built-In Email MCP Server 08:02 Testing Poker Agent with MCP Tools 10:35 Analyzing Agent Run History & Logs 11:27 Optimizing Costs with Allowed Tools Feature 13:10 Benefits of Abstracting Tools to MCP Servers #AzureLogicApps #MCPServers #ModelContextProtocol #AIAgents #AgentLoops