DEV Community

Saras Growth Space
Saras Growth Space

Posted on

What MCP Actually Is (Without the Jargon)

In the previous post, we saw the problem:

LLMs are powerful, but they can’t interact with real systems on their own.

So the question becomes:

How do we safely connect a model to tools like APIs, databases, or files?

That’s where MCP comes in.


🧠 The Simple Definition

MCP (Model Context Protocol) is:

A standard way for an LLM to discover and use external tools.

That’s it.

But let’s make this more intuitive.


🔌 Think of MCP Like a Universal Adapter

Imagine every device had a different charging port.

You’d need:

  • different cables
  • custom connectors
  • messy setups

Now imagine a single standard like USB.

👉 That’s what MCP does for LLMs.

Instead of custom integrations for every tool,
you get a standard way to connect everything.


🧩 The 3 Core Pieces

To understand MCP, you only need to know three components:


1. The Model (LLM)

  • Understands user input
  • Decides what needs to be done
  • Chooses which tool to use

👉 It is the decision-maker


2. MCP Client

  • Talks to the model
  • Sends available tools to it
  • Reads the model’s decision
  • Sends requests to the server

👉 It is the coordinator


3. MCP Server

  • Exposes tools
  • Executes real logic
  • Returns results

👉 It is the execution layer


🔄 How They Work Together

Let’s walk through a simple example.

User asks:

“Show my recent orders”


Step 1 — Client prepares context

It sends the model:

  • the user query
  • list of available tools

Step 2 — Model decides

It thinks:

“I should use get_user_orders


Step 3 — Client sends request

The client takes that decision and sends it to the MCP server.


Step 4 — Server executes

  • Fetches data from database
  • Returns result

Step 5 — Model responds

The result is passed back to the model,
which turns it into a human-readable answer.


🧠 Important Clarification

The model does not:

  • call APIs
  • run code
  • access databases

It only:

decides what should be done

Everything else is handled by the system around it.


🔥 Why This Is Powerful

This separation gives you:

  • Cleaner architecture
  • Reusable integrations
  • Easier scaling
  • Less hardcoded logic

Instead of writing rules for every case,
you let the model choose actions dynamically.


🧭 A Better Mental Model

Think of it like a team:

  • Model → decides what to do
  • Client → communicates and coordinates
  • Server → actually does the work

🧭 What’s Next

Now that we understand what MCP is, the next step is:

What exactly are these “tools” the model uses?

That’s where things start getting really interesting —
because tool design directly affects how smart your system behaves.

Top comments (0)