All videos
0:00 / 0:00
ai

Building your own MCP server | Made for Dev Ep. 5

Global AI Community16 June 2026Watch on YouTube

Part of series

Ep. 4 · MCP Servers & Protocol

Uitleg en praktische implementatie van Model Context Protocol voor AI-ontwikkelaars.

View the series

Description

In this fifth episode of the Made for Dev Show, host SIMON is joined by Stuart, a Microsoft MVP and Software Engineer, for a comprehensive guide on building your own Model Context Protocol (MCP) server from scratch using the C# SDK. 👉 Core Topics Covered 1. What is MCP?: A brief overview of the open-source standard for connecting AI applications to external systems like APIs, databases, and local software [00:34]. 2. The Three Building Blocks of MCP: • Tools: Functions that the LLM can execute (e.g., "search flights" or "send messages") [01:45]. • Resources: Read-only data sources that provide context to the agent (e.g., database tables or binary files) [02:08]. • Prompts: Reusable templates or examples that help users interact with the LLM more efficiently [02:41]. 3. Hands-on with C# SDK: Stuart demonstrates how to build a Dunder Mifflin-themed MCP server using the C# SDK (currently in preview). He shows how to use reflection and attributes ([McpServerTool]) to expose functions to the LLM [04:17], [07:14]. 4. Local vs. Remote Servers: • Local: Uses Standard IO (stdio) for communication within your machine [05:52]. • Remote: Uses HTTP transport, allowing you to host the server in the cloud (e.g., Azure) [06:18], [10:23]. 5 .Integration with Developer Tools: • GitHub Copilot: Connecting a local MCP server to VS Code's Copilot Chat to query a live database [11:10]. • Azure AI Foundry: Deploying the server to the cloud and integrating it into a custom AI agent [19:27]. 👉Key Highlights • Dunder Mifflin Demo: See the server in action as it retrieves employee details for Jim Halpert directly from a database through a simple natural language query [12:26]. • Prompt Templates: A demonstration of how slash commands in VS Code can pre-fill complex prompt templates for the user [14:46]. • Employee Handbook Resource: How to attach an entire handbook as a context resource so the LLM can answer questions about "company values" with zero external training [18:01]. #GlobalAICommunity #MadeForDev #MCP #ModelContextProtocol #DotNet #CSharp #AIAgents #SoftwareEngineering #AzureAIFoundry #GitHubCopilot

What you'll learn

  • MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems like APIs, databases, and local software.
  • MCP servers have three core components: Tools (functions the LLM executes), Resources (read-only data sources for context), and Prompts (reusable templates).
  • Using the C# SDK you can build MCP servers with reflection and attributes ([McpServerTool]) to expose functions to the LLM.
  • Local servers use Standard IO (stdio) for communication on your machine, Remote servers use HTTP and can run in the cloud like Azure.
  • MCP servers integrate directly with GitHub Copilot in VS Code and with Azure AI Foundry for cloud-based AI agents.

Frequently asked questions

What are the three building blocks of an MCP server?
The three building blocks are Tools (functions the LLM can execute), Resources (read-only data sources that provide context), and Prompts (reusable templates for efficient LLM interaction).
What is the difference between Local and Remote MCP servers?
Local servers use Standard IO (stdio) for communication within your machine, while Remote servers use HTTP transport and can run in the cloud, such as on Azure.
How do you integrate an MCP server with GitHub Copilot?
You connect a local MCP server to VS Code's Copilot Chat, enabling it to query live databases and other external systems through natural language.
How do you use resources in MCP to provide employee handbooks?
You attach a complete handbook as a context resource so the LLM can answer questions about company values without requiring external training.

Topics

mcpai-agentsdeveloper-toolsc#azure