MCP & why API's don't get vibes

There is a lot of confusion about MCP (Model Context Protocol), especially when compared to the traditional API approach.What is MCP anyway, and why don't APIs get vibes?

So grab your drink and let me take you to a party. 

As the night goes on, you approach the DJ booth. The DJ, a shiny robot is taking the song request from you.

But with its fixed playlist and rigid programming, the old-school robot doesn’t get your vague, high-level request.

We give it another try. "Play whiskey in the jar"!And even with this particular song request, it doesn't work.

No Context, no fallback, no fun…

The guests leave the party because the music selection is too limited, and they’re not enjoying themselves.

The party ends with unhappy guests and an empty dance floor…

The Broke DJ

The party ends with unhappy guests and an empty dance floor…

The DJ lost its reputation. His old, tightly coupled CD collection API setup feels outdated now. And the crowd wants fresh mixtapes...

To become relevant again and rise back to star DJ status, he needs access to more sources he can play from.

But every new integration costs $$$ – and he’s already broke.

MCP? MCP!!!

The DJ walks home, broke and frustrated, carrying his old controller.

It stumbles across something new. MCP? MCP!!!MCP. A growing directory of tools. Each one ready to work out of the box. Each one following the same rules.

Our DJ Robot Agent doesn’t need to rebuild everything from scratch. Before, every new source meant rewriting the whole setup. And the DJ couldn’t afford that anymore anyway. With MCP, there’s no custom wiring for each tool. No guessing what a service can or can’t do. Every tool speaks the same language. He just connects and plays.

The MCP Comeback Set

Now, equipped with MCP the DJ is back. You repeat, "Can you play something to cheer me up?" This robot pauses for a moment as it processes your request in a more human way.

It understands you're asking for an upbeat, mood-lifting song (even though you didn’t name a specific track). Behind the scenes, it’s using MCP to access multiple music sources.Because MCP is connected to external services it responds: "Sure! "How about this song …" and starts playing it.

The MCP DJ robot is able to interpret a prompt, figure out what you really want, look up options across different sources, and then choose an appropriate response (the song) to satisfy your request. It’s not limited to one playlist or one set of instructions - it can mix and match tools (search, databases, etc.) to deliver what you asked for.

Putting all together

Let’s break this down with a real actual MCP architecture.

MCP Host + Client

On the left, we have the MCP Host with its MCP Client (likely Cursor, Claude, ...). The client receives a user prompt "play some EDM" and then decides which tools to invoke (e.g. "playTrackYoutube") 

MCP Server

Each MCP Server connects to a specific data source or service, like Spotify API, Soundcloud API, ...

These services handle the actual work: searching for songs, playing tracks, and more. The MCP Server acts as a translator, converting standard MCP requests into API calls that the specific service understands.

The Advantage

No glue code. No weird formats. Just one shared transport for every tool (JSON-RPC 2.0)

And that’s the point of MCP. It lets the model use tools on the fly. Your AI assistant becomes a flexible operator. It picks up whatever tool is available, as long as it follows the same playbook.

On the fly means you can swap YouTube for SoundCloud. Or just add YouTube as another source.

Wait! Wait!

Some might argue that you could build the same functionality with an AI Agent + traditional APIs. Thats true! But the point of MCP as a Standard isn’t that it suddenly unlocks something we couldn’t do before to be honest.

The difference is how much work it takes to make that agent actually useful across tools. Traditional API setups often need custom logic, specific endpoints, and hardcoded instructions for each service.

This means developers must write and maintain separate code for every tool or data source the AI agent interacts with

"You’re basically hand-coding every skill the agent has."

With MCP, you skip most of that. The protocol defines how tools should describe their capabilities, so the AI can figure things out on its own, in real time. It’s more like giving the agent a shared manual for every tool, instead of making it memorize a new one each time.

The End

And our DJ?

It can play whatever the crowd wants. He plugs his services in, and it just works. Same interface. Same protocol. No custom fixes. Swap SoundCloud for YouTube. Add something local. Or mix in all of them. He just shows up, connects, and plays.

Both approaches (API + MCP) works. But MCP removes the busywork. And gives our DJ a shot at the comeback he deserves.