Contextualizing MCP Servers and Clients in Agents

00:00 This will be quite a long journey. An interesting one, but a long one. So let’s make sure you understand what everything is and what everything does. Remember that an MCP server provides prompts, resources, and tools to be used.

00:16 Now, the server only exposes them. An MCP client is anything that can connect to an MCP server to request prompts, resources, and tools. In this course, you’ll write one MCP client.

00:30 You are going to write one MCP client that will be able to connect to any MCP server, and to interact with it deterministically. You specify which subcommands to run, which might just connect to the server, it might list the members from the server, or it might get a specific prompt, fetch a specific resource, or call a specific tool with specific arguments.

00:56 And this is deterministic, and you use your CLI interface to do this. In general, agents and similar tools embed an MCP client so that they can connect to MCP servers, but it’s the output of the LLM that determines whether or not it should use something from the server. So there is a clear distinction between the way you’re using your MCP client through the CLI and the way LLMs use their MCP clients.

01:30 This is a major distinction. If you were to implement your own agent, for example, you would also need to write an MCP client, but you would use it in a slightly different way because it would be the LLM driving when and how to connect to the server.

01:47 Hopefully this makes sense. So when you’re ready, move on to the next lesson where you’ll learn about the prerequisites for this course.