Orchestrating Business Logic With AI Agents: The Power of MCP

Introduction

Yeah, so hi everyone.

What Is MCP? A High-Level Overview

Yeah, I gonna to present today a little bit technical and more demo session for the interesting topic for everyone is for MCP. It's model context protocol, which is kind of like communication protocol between agents. So we have a, like I will explain like, so we have a website which talking about HTTP, if you know, and MCP is protocol which agents is talking, yeah?

Speaker Background

So about me, my name is Dmitry. I'm a software developer from Apple.

I work here like three years and I work in basically an LLM topic.

Understanding MCP

So what is actually MCPs? If no one knows.

Why MCP Matters for Agents and Tools

Simply talking, it's a protocol that helps you to connect a lot of different tools. Imagine that you have a remote service like Slack, Google Cloud, Calendar. You can name any service that actually you're interested in.

Open-Source Origins and Ecosystem

Because currently it's open source, it's created by Anthropic, those people who created Cloudy, gave us Anthropic models.

Real-World Use Cases and Integrations

and they open sourced it and it's get like a very huge impact like in terms of communication like between services simply it's like you have for example you have your chat gpt and you want it to talk with your Google Docs, you want to talk it with your calendar, you want to talk it with maybe your WhatsApp, Telegram, whatever you want, you can easily use this service, connect to it, and you can ask ChatGP, okay, go to Telegram, create this chat, and send it to Gmail and send it to maybe to your boss, yeah? And it will do it for you because of the access like to each of these services, yeah?

Why Do We Need It?

Um, why do we need it?

Yeah.

LLMs Need Tools to Be Useful

So LLMs is like, they are not, they're kind of dump, honestly, like all this LLMs raw, they don't know anything like, but if you give them a lot of tools, it can do a lot of interesting stuff.

From Understanding to Action

So LLM with such thing, like can not only like they can, um, understand the world around it. Around, uh, what you show.

And I can, I later, I will show you.

Yeah. Yeah. And, uh,

Coordinating Tools to Achieve Goals

1With MCP, you can easily adapt, you can coordinate with different tools, and it can achieve easy goals, basically.

Demo Setup

So let me maybe, because my demo is more practical, so I will show you actually what I prepared for you.

Problem Framing: Generating Business Ideas

Just a question to you, anyone tried to generate a business idea? Okay, we have such people.

I also struggle with this because I'm not a very business guy, but I want to be someday.

So imagine that you want to get a business idea. You don't know actually where to start. You're thinking, okay, there's some problem, but I don't know how to put it in my mind.

Mining Reddit for Pain Points

What with LLM you can do, anyone knows Reddit? Yeah, anyone use Reddit? Cool.

So what we can do here is we can go to Reddit, to subreddits, and ask it, okay, give me maybe parse the posts and give me back ideas that actually people want to solve. And in LLM, you ask it, okay, try to find the comments of people which, for example, oh, I want to solve this problem.

I want to pay for this problem. This product is very bad. I want another one. And you try to find this kind of posts which actually gives you hints actually where to start.

So, for example,

Data Pipeline: Sheets, Triggers, and Models

I have this flow, I have a Google Sheet, which is empty, and then I can trigger execution. So what I do, I have three subreddits inside.

I don't remember which post, it's like some internship, internship, lunch product. whatever and what they do it's merges it and I use Google Germany to analyze all the posts that actually it found so I took from each from each year post like three three of which because Germany is not very expensive but it gives you a credit quote you can use it like 50 requests or per day yeah and

Model Choice and Rationale

but i i uh yeah you can use any model that you want but gemini i from my opinion this works for best for now and it's have very good integrations because um honestly open ai started the drive but google now trying to catch them yeah and you see like now it's created like i can show you here what happened so here you have a

I know it's fine.

Inside the Prompt and Run

So this is what's happened inside of the model. There is a prompt. There is a model.

I use a very simple Gemini 1.5. There is more advanced and more expensive. You can get very good results. But I took the one that actually is cheap.

And it's finished the process. And here you see that actually some text appeared.

Sample Outputs and Insights

So what it did, it's like we are interested in output. So here he said, okay, uh, let me check. Uh, yeah.

For example, startup idea, verify your gig, a plan for, for refined side hustle claims. Uh, another startup idea, Etsy launchpad, uh, another idea, Genesis project, like, and like it's explains actually what is the focus?

What is your target audience? It's gives you like a, what is a Reddit link from way that good took.

I don't know, decentralized, geo-targeted, volatile platform. So all these conclusions came from some people's comments. They struggled, probably. They wanted to pay or they're interested in, yeah?

Where MCP Fits in This Workflow

So, under the hood, what happened here, this is not MCP yet. It's just, like, workflow that I showed you in here, yeah? MCP will be happening here.

But this is, like, how you can get a lot of, like, traction of the ideas, like, of people wanted to solve, yeah?

Q&A Highlights

And then... I have a question. Sorry. Yeah, yeah. How did you get it?

How Were Posts Selected?

Did you choose the three posts? No, subreddits. But you went into the subreddits and picked out the posts that you wanted to analyze.

It didn't look for you. No, no, it looked for me. It parsed it.

So you see here? What was the original directive you gave it?

Subhustle. OK. Side hustle.

I just entered inside of it. I took it top.

And then I just took... Within Reddit, though? Yeah, within Reddit. Okay, got it. Yeah.

From Ideas to Execution with MCP

And then, like, for example, you said, okay, I have an idea how to execute it.

So I go, like, to the MCP target. I start it.

Connecting Tools: Cursor and n8n

And then everyone knows Cursor. You, like, have a connection between the Cursor now and the NS3 instance. And you say, okay... Sorry, everyone know what is Cursor?

What Is Cursor? (Clarification)

No. Okay. So Cursor...

Yeah, Coursera is one of the tools that actually you can create your website just like typing, OK, I have this idea. Please create for me a website. And it will create it for you, basically.

It's a tool that actually helps you to create your online websites. It's the simplest way. But it can create something more complex.

Automating Selection of Best Ideas

So you say, OK, inside n8n, find me best idea. And show it.

Yeah, like Coursera is like basically there is a couple of tools which is Coursera is one of these that actually can help you to create your websites, platforms, whatever. And it's this one I use. It's open source.

Cost and Practical Considerations

But if you want something like more advanced, you have like $20 per month. But it's very good. I mean, it's worth it.

And it's like, yeah, I'm just, how much time is left? Jorge, how much time is left? Oh, five minutes. Five minutes.

Six, don't worry. Six, OK.

MCP in Action: Reading and Ranking Ideas

Yeah, so you see, now it's taken all the inputs that I gave him from that file. Basically, this is where MCP is hidden. So my cursor going inside my document, reading it,

and then it will make a judgment basically yeah actually what is inside so it's still thinking yeah so now it started to do it i pulled that yes from your google sheet and rank it up for best idea vote is uh digital product sell problem solving digital products inspection pdf guides start by picking an issue can generally help make a strong offer with quality guide blah blah blah yeah and it's linked to the reddit basically

Actually, we can check what is inside the base offline. Yeah. So he took this comment from top and then he analyzed it and then like he made the statement.

From Analysis to Prototyping

And then what you can do, like the last step, but I didn't try it. But it's a free option I have, probably if I have a paid one.

Auto-Generating a Website

Okay. Create a website. Website.

Sorry. Website for me.

Simple using HTML, CSS.

And it probably will start to do something. And it starts to create.

Live Build and Preview

I'm going to create index.html. Probably I can even open it for you if it will finish it.

Developer Workflow Tips

So I mean, I showed you the workflow, how you can. except yeah yeah yeah because cursor is is nice for developers like me for example because you can actually every time review the step so the idea yeah so i don't bother you very technically

Yeah?

Key Takeaways

You don't know what to do. You don't have idea.

Process Recap

1You can easily fetch ideas using LLM. And you can use MCP to talk with different services. And then you can create it instantly.

Speed to Prototype

And it will take you, if you're very good understanding the flows, you can make it in a couple hours. You can create a prototype.

Iterate Quickly

Is it works? No. Trash it. Do it again.

No-Code to Low-Code Entrepreneurship

And this is actually how people are doing now, because you don't need to be a developer to create your own business in the web.

Results and Reality Checks

You see, it's created some kind of HTML. And you need to run it, I don't know.

Okay, so there's some parsing error, but usually after some time it's trying to recover it and then finish.

Conclusion and Next Steps

Yeah, I think you have a lot of questions, so I will stop here.

Finished reading?