AI Agent Mastery Certification Course: Module 4 – Tools & MCP

[00:00:00]Welcome back. In this module, we're going to go ahead and look at tools and MCP. Um, these are both basically kind of what lets agents actually do things beyond just generating text. So, these are the concepts that really open the door to real world um actions like querying data or running code or in general, I think if we took a step back, just generally connecting to external systems.

[00:00:24]So this module is kind of all about understanding um how tools extend an agent's abilities and on the MCP part MCP part of it at least it's how MCP kind of creates a simple yet kind of consistent way for agents to actually interact with external services or even your local resources and I feel like in the last couple modules we've been talking a lot about tools um talking about tool spans talking about um different you know agent workflows and stuff that have tools but we never actually defined what is an AI agent tool. So agent tools in short are kind of external capabilities that the agent can call when it needs information or essentially needs to take some sort of action. So on this slide you can go ahead and see categories like um data access, extraction, analysis as well as execution. And I think that the benefits here are pretty clear to see.

[00:01:16]In general, tools give grounded knowledge and let your agents uh take real actions to make your systems both modular as well as scalable. Um, and so in short, tools kind of help agents go far beyond what a base LLM can do on its own. So even things like looking up things on the web or um calling APIs or anything, those are all u properties of tools and you need a tool to be able to do those things.

[00:01:42]And so why do tools matter? because tools kind of extend uh the model's capabilities like we said. So they let the agent kind of fill in knowledge gaps using external data and they also kind of allow some amount of automation.

[00:01:54]Automation meaning like things like code generation or data queries or planning or like I said in the past web searches and so the examples at the bottom of the slide here that kind of show the range of which um of what tools can really trigger. And I think that without these tool calls, an agent is kind of limited to just generating text because at that point you are just an LLM, right? That's like the only part of your agent that you have other than tools really. Um or the other parts of your agent help you call those tools, right? And so if you were to just have LLMs, then you just have something that is able to generate text for you.

[00:02:31]So now moving into um MCP. um MCP or model context protocol, that's what MCP stands for, is a standard way for agents to kind of talk to um local or even remote services. So I think this diagram um kind of shows how a host host meaning things like cursor or cloud code or cloud desktop, right? And how that connects to multiple MCP servers. So each of these MCP servers um exposes some sort of capability. So something like uh one could be accessing an SQL um database, right? Another one could be deploying an app, another one could be doing something simpler like sending an email, right? And so the agent essentially interacts with these services through the MCP client. Um which means you can plug in new capabilities without actually having to change the agent itself. And so it basically creates kind of a clean and flexible way to connect your agent to both your computer as well as the internet.

[00:03:28]And that was pretty much the base definition for um what MCP is. And so we'll go ahead and start talking about the lab a little bit. Um in this lab, you'll kind of compare how your agent behaves when using regular tools versus if you uh if your agent were to start employing MCP. So you'll be able to kind of see the differences in things like Rag and MCPdriven actions. And then from the last module when you did your orchestrator worker workflow um for your orchestrator agents you'll go ahead and also look at how the agent chooses which sub aent to call and whether it is actually able to preserve the right context um as it's doing that. And so questions like these kind of help you evaluate if your agent is actually even reasoning correctly.

[00:04:11]And so now we're going on to the hands-on part like we can go ahead and head to um lab 4. Um and then there you you'll go ahead and implement tools and connect your MCP servers um just so that your agent can pull data, run operations, and kind of perform more grounded tasks. Um like always, you can go ahead and start working through the notebook on your own or you can go ahead and watch the lab walkthrough. In general, regardless, it'll kind of give you some sort of practical um experience wiring up real capabilities into your agent.

[00:04:43]That's pretty much it for what we had for this module. It's kind of on the shorter side, but I think the lab will do a lot more of the learning part for you. So, I'll see you guys all in the next one.