Try Obot
Request a Demo
Tutorials

What are MCPs? A Simple Introduction

May 28, 2025 by stein ove helset

MCP or Model Context Protocol is a relatively new concept in the world of AI, but what is it and how can it help you?

AI is constantly improving, and becoming more and more advanced. Learning and understanding every concepts can be really hard, and there is a lot of information out there.

In this article, I’m going to break down the concept of MCP, and hopefully keep it short enough for you to stick around untill the end.

What is MCP?

Well, let’s begin with AI assistants. They are smart, but they have a big flaw. There are no good ways to handle history: connecting to a system like Slack or Git requires an integration built from scratch, etc.

And that is where MCP is coming in to the picture. With MPC, we now have a much simpler way to get the data we want without reinventing the wheel over and over.

We can get real-time access to relevant context, easily connect to any data source, and it’s easier to scale across teams, tools, etc.

So instead of building everything from scratch, we now have an open standard we can follow, making this much easier.

You can think of MCPs as a bridge between AI and the information from real-world systems. MCP is a concept created by Anthropic, the people behind Claude. They wrote an article when they open-sourced and announced it.

A MCP server can access both local data sources, as well as remote. This makes is possible to have extra security, or you can share your MCP servers with other users. And when we’re talking about security, with a MCP server, you will have clear guidelines to how authentication should be implemented.

Originally, a MCP server was designed to work on the client side through studio processes (Standard I/O processes), but the plan forward is to push it in a direction of Streaming HTTP.

A simple example

Let’s say you are planning a big road trip through Europe or America. For this you would probably have a lot of spreadsheets, schedules, wish lists, and similar in your Google Drive. You open up your AI assistant and ask:

“Can you find all the hotels we want to book, and see if we have forgotten to book any of them?”

Without MCP, the AI can’t actually look inside your Google Drive. It would just tell you something general like, “Please check your hotel list.”

But with an MCP, the AI could securely connect to your Google Drive (Using the open MCP standard), find your spreadsheet with all of the hotels, and give you an answer like:

“You have forgotten to book on these 2 hotels: Hilton, New York and Caesars Palace, Las Vegas”.

As you can see, the MCP now did all of the work for you because it can access all the information it needs, and it also knows exactly how to respond.

Diving deeper into MCPs

  1. MCP at the core
    You probably already know other protocols like HTTP, FTP, or SMTP. They all have they’re own purpose, and are designed to follow specific ways to handle different scenarios, how to respond, etc. And that is exactly what MCP does for AI
  2. MCP Server
    The “bridge” between the AI and the real-world information is the MCP server. This can be run locally, or at any server in the world. You can have one MCP server exposing files from your Google Drive like photos, or documents. You can even have it expose the contents of your documents in a way that’s optimized for the AI to understand.
  3. MCP Client
    This is what we use to talk to the MCP server. For example, the chat in Obot can be a client, or Claude running in your browser.
  4. Specifications
    This is a list with definitions on how to request and respond between the server and client. It can be information about the authentication and permissions. And which operations can be used on the server like “list”, “read”, “write”, etc. It also includes specifications for prompts, resources like files, other types of contents, etc. This makes it easier to get more consistent answers from the AI.
  5. Pre-built connectors (or MCP servers)
    There are a lot of pre-built connectors out there you can just install and start using. These are very easy to use, and you don’t have to do anything besides installing them and do some configuration. There are MCP servers for Slack, GitHub, Google Drive, and a lot more. Here’s a list of example servers you can play around with.
  6. Security
    A very important aspect of MCP servers are the security. Instead of just giving AI access to your whole Google Drive, you can use the MCP server to control the content it gets access to. And also if it can only read, or also write. MCPs use standard auth methods so make sure you’re data and connection are secure and private.

How do I get started?

It’s not always easy to know where you should go to get started with a new concept like this. I would recommend checkout out this website (https://modelcontextprotocol.io/introduction) where you can find a lot of good information on how to install and set up your first MCP server. You will also find SDKs for a variety of programming languages like Python, TypeScript, Java, and Kotlin.

I will go through how to do this using Python since this is what I’m most familiar with.

  1. Make sure I have a new version of Python. I’m using 3.12 (You need 3.10 or newer).
  2. Install and set up an environment. I’m going to use a Python project called “uv”. This needs to be installed, then initialize a virtual environment and install the dependencies.
  3. Creating the server it self is not necessarily too complicated. I’m going to use a free API (https://kinduff.github.io/dog-api/). I will not be going in to details for the script, but you are going to need a function for querying the API and format it correctly when you get it.
  4. Next, you will need a function you can call from the MCP client. This functionswill get the request from the client, call the api, and return the data exactly how we want to return it.
  5. Last but not least, we need to run the server.

When the server is running, you should be ready to test your MCP server from a client.

For even more details on this Python guide, go to this quickstart for Python.

Resources

What’s next?

Now that you have learned the basics of MCPs and how they work, you can try to actually use a pre-built MCP server. Or if you’re feeling up to it, try building one. If you want to set up a cloud based MCP-connected agent, Obot.ai provides a simple approach to creating one for free.

By just going through setting up something like this a couple of times, you will probably start understanding how much the MCPs can help up with the correct information.

Related Articles