MCP: AI Apps as Easy as 1, 2, 3

1. MCP: AI Apps as Easy as 1, 2, 3

Hi, and welcome to this course on MCP!

2. Meet Your Instructor

This course was created with Korey Stegared-Pace, Senior AI Cloud Advocate at Microsoft, and instructor of Microsoft's Generative AI for Beginners and AI Agents for Beginners courses.

3. Looking Ahead

In this course, you’ll learn how to bridge the gap between LLMs and the real world and build AI for scale. We’ll explore how the Model Context Protocol, or MCP, enables smooth and efficient integration between AI systems and data sources, tools, and workflows. By the end, you’ll understand how MCP works, how to build your own MCP servers from scratch, and connect them to LLMs.

4. The Big LLM Problem

Large Language Models, or LLMs, are incredibly powerful, but have one major limitation: they can’t access real-time or external data; they're aware only of the information they were exposed to during training.

5. The Big LLM Problem

Think about the information you check on your phone every day—sports scores, traffic updates, emails, and more. The apps you use access that information via

6. The Big LLM Problem

APIs, which developers build using different access controls, endpoints, and data formats. When LLMs began using APIs to access similar data and extend their capabilities,

7. The Big LLM Problem

the integrations quickly became complex. Each tool or API required custom code to connect it with the AI system.

8. MCP to the Rescue!

To solve this problem, the AI industry needed a single, standardized way for AI systems to access external capabilities. That’s why the Model Context Protocol, or MCP, was introduced.

9. MCP to the Rescue!

Often described as the “USB-C port for AI applications,” MCP simplifies connectivity and adds features that traditional APIs lack, such as dynamic tool discovery. This means the LLM can see which tools an MCP server offers in real time while completing a request. Before we dive into how this works in practice, let’s look at the three main components that make up the MCP architecture: the Host, the Client, and the Server.

10. The MCP Architecture

The host is the AI application the user interacts with directly. This is where the connection to an MCP server begins. The host receives the user’s query, and based on that query, the LLM determines which external connections it needs to make.

11. The MCP Architecture

MCP servers are systems that provide AI-friendly access to external data or functionality, and the host makes requests to a server to access its data or functionality. These servers can run locally on your machine or be hosted in the cloud.

12. The MCP Architecture

They essentially translate complex API or database calls, each written with their own protocols, into the consistent, easy-to-use MCP format.

13. The MCP Architecture

To establish that connection, the host relies on clients. The client is responsible for creating a session with the MCP server

14. The MCP Architecture

using a structured message format called JSON-RPC. During this session, the client

15. The MCP Architecture

receives tool definitions—names and descriptions—from the server. The LLM processes these definitions, decides which tools are needed, and the client executes the corresponding requests. The server then returns the results, which are passed back to the LLM to include in its response.

16. Primitives

MCP servers offer functionality in the form of primitives. The first primitive, tools, refers to the actions the LLM can perform to complete a user’s request, such as querying a database, booking a flight, or running code. The client triggers these tools, and the server executes them before returning the results to the LLM.

17. Primitives

The second primitive, resources, provides read-only data or context that the AI can use. This includes documents, files, knowledge base entries, or database records.

18. Primitives

Finally, prompts are predefined workflow templates or sets of instructions that guide the AI through specific tasks. They save users from writing long, complex prompts and ensure that the model follows best practices.

19. The MCP Architecture

Together, these three components—host, client, and server—allow MCP to provide a unified and flexible framework for AI applications to connect to tools and data

20. Let's practice!

safely and efficiently.

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.