0:01Say you're building an AI assistant for your team. It chats, it answers questions, and people actually use it.
0:07Now you want it to do real work, pull up an issue, check a commit. That means talking to GitHub.
0:12So you connect them. The model gets a function it can call, and GitHub's API is at the other end.
0:17That's a complete round trip, the assistant asks, and GitHub answers.
0:21That's plain function calling, and for one app talking to one tool, it's all you need, no standard involved.
0:28That connection isn't free, though. Somebody wrote the sign-in flow, converted the data, caught the failures, and now maintains all of it by hand.
0:36One app, one tool, one piece of glue. Keep that picture in mind, it doesn't stay this simple.
0:43Now that your assistant is useful, people want more from it. They want it to read from the database, and post updates to Slack.
0:50Each of those is another integration, with its own sign-in, data conversion, and failure handling, written just for this pairing.
0:57Three tools, three custom integrations. Still manageable.
1:00Then the rest of the company shows up. The coding agent in your IDE needs these same three tools. So does the support bot.
1:07And none of that glue is shared. Every app rewrites every integration, from scratch.
1:12Three apps, three tools, nine separate integrations to build, test, and maintain.
1:17And it doesn't stop at nine. Dozens of apps, hundreds of tools, every pairing is one more integration. That's the N by M problem.
1:25MCP is the fix. Anthropic published it as an open standard in late twenty twenty-four, and the industry quickly adopted it.
1:32This is MCP, one standard, between the apps and the tools. Instead of every app implementing every tool's private interface, they all agree on one shared way to talk.
1:42With everyone on the same standard, all that custom glue can go. All nine hand-written integrations.
1:49Each app now implements the protocol once, no matter how many tools come later.
1:53And each tool needs just one MCP server, which every app can use.
1:57So each tool now has its MCP server. That server is a small program that speaks the standard on one side, and calls the original API on the other.
2:06Count the connections again. Six instead of nine, and the savings grow with every app you add.
2:11The cost of adding something new now grows by addition, not multiplication. That saving is the reason MCP exists.
2:19And traffic still flows, the app asks through the standard, the answer comes back.
2:24So far, we've only called it by its short name.
2:26MCP stands for Model Context Protocol.
2:29That's the standard you just watched at work, and there's more to it than one video can cover. So this is the first video of a series.
2:37The series has three sections. Let's look at them.
2:40The first section is the idea.
2:42We start with the name, and what each of its three words means.
2:45Then who the host, the client, and the server are.
2:48Then what a server can offer, and we plug a real one in and use it.
2:52The second section is the build.
2:54We write the smallest possible server.
2:56Then we look at the messages between it and a client.
2:59And then an agent gets access, and discovers the tools at runtime.
3:02The third section is using it for real.
3:04That means trusting a server the model acts through.
3:07Then wrapping a real service, and moving the server off your machine.
3:11Next time, we start with the name itself.
3:13Each of its three words says something about what MCP is for, and that's the next video.

What Is MCP, and Why Do We Need It?

Created by Abhyuday DesaiPublished 2 Oct, v1Runs 3 min 25 sec

3 views

About this video

In Episode 01 of MCP Explained, we start where most teams start: one AI assistant, one tool, and a hand-written integration between them. Then the team wants more tools and the company wants more apps, and the integrations multiply into the N by M problem. MCP is the fix: one shared standard between apps and tools, so each side implements it once and every tool becomes an MCP server.

You'll learn how to:

See why plain function calling is all you need for one app and one tool, and what the glue around it costs to maintain
Follow the integration count as apps and tools multiply: three by three becomes nine, and it does not stop there
Understand how one standard between the two sides turns N times M into N plus M
See each tool become an MCP server: a small program that speaks the standard on one side and calls the original API on the other
Learn what the letters stand for, and what the three sections of this series cover: the idea, building and connecting a server, and using MCP in the real world

This video is for developers hearing about MCP for the first time. No prior MCP knowledge is needed.

What changed

v102 Oct 2026abhydesai3 min 25 sec805b43c

First published version.

3 views1 like on this version