Diving Into MCP Servers and a Workflow Example

This lesson preview is part of the MCP in Practice: The Future of AI Agents course and can be unlocked immediately with a \newline Pro subscription or a single-time purchase. Already have access to this course? Log in here.

This video is available to students only
Unlock This Course

Get unlimited access to MCP in Practice: The Future of AI Agents, plus 70+ \newline books, guides and courses with the \newline Pro subscription.

Thumbnail for the \newline course MCP in Practice: The Future of AI Agents
  • [00:00 - 00:05] And if we talk about server, it's not that complicated as well. Once again, guys, let's just go step by step a little bit deeper, deeper.

    [00:06 - 00:18] And just in general, over you, what is going on on the server side, we have three main parts, primitives of features. Like from now on, we are going to change, use it interchangeably.

    [00:19 - 00:24] Those words, we don't get so don't get confused. And those are tools, resources and prints.

    [00:25 - 00:27] Okay. What are they?

    [00:28 - 00:31] Let's start from the tools. You're going to use them most of the time, actually.

    [00:32 - 00:39] Tools are just for doing action. If, for example, you're a web developer, it's just like you're making a post request, right?

    [00:40 - 00:50] And you're asked for some data to be maybe fetched or processed and something like that. So we are doing, we are using tools actually to do the job for us to do the task for us.

    [00:51 - 00:55] Okay. Resources, on the other hand, is just forgetting data.

    [00:56 - 01:03] So we can get data into context of our element of choice. It's just, it is like a get request, right?

    [01:04 - 01:10] We want you just to get some data and that's it. And yes, actually we can use tools to get the data as well.

    [01:11 - 01:18] You might still watch the difference. And actually in terms of the resources, we can manually enable it in our host side.

    [01:19 - 01:25] So we do not need to ask questions like, Hey, go grab something to do some task for a mirror or something like that. Right.

    [01:26 - 01:35] We're just enabling it and it automatically fetches all the data and it always available for our LLM to use. So we don't need to worry about it.

    [01:36 - 01:41] And so we can simply use tools, prompts and other stuff without worrying that we are liking some context. Okay.

    [01:42 - 01:49] And then we have plans. Bronze are actually just the templates that we are using, like prebuilt bronze, basically, right?

    [01:50 - 02:05] Just in the name, which we are using to describe what do we want and what exactly what pieces information we want to send right now. As an example of the cloud, we would talk about a little bit later on, but this is just like a chat GPT you can think.

    [02:06 - 02:06] Okay. If you're not familiar.

    [02:07 - 02:20] And on the cloud sides, when we are using prompts, it will just give, will give us the input section, basically where we're inputting different data. For example, what you want to get data, whether from, okay?

    [02:21 - 02:35] And we are tapping like New York or something like that, right? And basically we're just hit sense and it already is like in the prompt, all the information about how we want maybe the data back, what details about whether we are interested in.

    [02:36 - 02:49] So it's basically just template for us to use, or maybe you have a large team which uses the same prompt again and again, it's just easy way to manage this and actually prepare everything beforehand. And those are three main parts.

    [02:50 - 02:56] Today, we're going to specifically use tools. I know because this is one of the most used parts in any client side.

    [02:57 - 03:04] But no worries. Once you figure out tools and you would see the example, you would understand like the way how to use resources and as well.

    [03:05 - 03:13] The most important idea just to have the understanding exactly like what is going on, right? And then you can use whatever you want and play around with it as much as you want.

    [03:14 - 03:23] But before going forward, let's not forget that actually once again, MCP is quite new, right? It's emergent technology.

    [03:24 - 03:32] And that's why in the first link that's in the second link that I have sent to you. I basically mentioned that we will go back to it and this is the time, right?

    [03:33 - 03:43] So while looking through the hosts that are available right now for us to use, right? You can see a bunch of checkmark or no, it's not used.

    [03:44 - 03:57] OK, one sec, I will to send the link, no issues. Yeah, just wait a little bit.

    [03:58 - 04:02] You guys should be able to see this kind of the table. Yeah, yeah, that's great.

    [04:03 - 04:12] And yeah, they have things guys. And basically you right now able to see that it's not really supported everywhere.

    [04:13 - 04:17] And specifically look at the tool. We have resources, prompts and tools sampling and routes.

    [04:18 - 04:21] Don't like stress about this for now. It's not almost used anywhere.

    [04:22 - 04:33] And those are important stuff for us to focus, especially at the beginning. And depending on the app that you're using, actually, if resources are important for you, you can use probably client, right?

    [04:34 - 04:40] Because it starts inside of the, for example, course, right? For example, prompts is somewhere is missing, right?

    [04:41 - 04:51] Tools are usually once again, since the tools are the most used. Right now, it's usually important in most of the cases because usually you want to perform some tasks or something like that, right?

    [04:52 - 05:01] Instead of fetching, even though this is super useful, but tools right now, the way to go. So when you're choosing the host for your needs, please take a look.

    [05:02 - 05:09] Also add the what is available right now and what is not because it might be breaker for you. OK.

    [05:10 - 05:17] But if we are like doing some classic approach, most of the clients actually would support. So keep this in mind, please.

    [05:18 - 05:36] And just to one more time, let's say, remember better the process, right? Let's imagine the full process right now, understanding everything that I've mentioned before about how everything is connected and about the workflow, which we are going when we are using the MCP.

    [05:37 - 05:41] OK. So first of all, oh, what are the next slides?

    [05:42 - 05:46] Sorry. So first of all, we are asking about the weather in your or sacrament somewhere else.

    [05:47 - 05:54] OK. We are asking actually MCP host and MCP first receives our very two process.

    [05:55 - 06:07] Then cost uses MCP clients, which is connected to our server or servers. If there are a lot of them, then client checks the server availability and prim itives.

    [06:08 - 06:12] Primitives is basically tools or something like something else, but usually just tools. OK.

    [06:13 - 06:21] And then we see that servers working is connected and it includes tools with the right description for us to use. What is happening?

    [06:22 - 06:31] It sends initial query plus information about this server back to the host LLM. So for now, we are not sure what to use.

    [06:32 - 06:35] So we are just grabbing information about the server. We have the initial query.

    [06:36 - 06:46] We're sending it back to the LLM and then a lamp having all these contexts can decide actually which server to use if it's a good one or not for the task. Then what we are doing.

    [06:47 - 06:49] What they can be? I don't just figure out what to use.

    [06:50 - 07:02] Then a lamp tells the host to go specific right server, which is actually choose before. And so we're doing a job and sends back a response.

    [07:03 - 07:12] And alone uses these two device click ads. So just to give you one minute, you can look through this path.

    [07:13 - 07:14] Great, guys. So let's keep it for moving there.