Articles

Apr 19, 2025

Reading Time: 8min

What Are Model Context Protocol MCP Servers and How to Use Them

Unlock the future of AI automation with the MCP Server and Modal Context Protocol. Discover how to future-proof your software with MCP integration and lead in the age of agentic AI.

Imagine this. You wake up one beautiful sunny day to your phone alarm telling you the weather, top tasks for you to do that day, and messages you received in the night. As you get ready for the day, you tell your phone to respond back to friends, colleagues, set up appointments, and meetings. All this before breakfast, without ever sitting down at a computer.
Have an idea for a marketing campaign? You don't have to hire a marketing agency and spend thousands on a campaign. All you gotta do is ask your phone assistant to generate a marketing campaign, have it write the copy and launch the campaign after you have reviewed it. Then the assistant analyzes the data and gives you the key insights and results in plain English.
Want to build a landing page? Just ask your assistant and it will create the copy, integrate with the necessary services, and launch it live.
Have paperwork to do? Your assistant fill out the forms and review before submitting it.
You even want to build a simple app? No problem. Just tell your assistant the desired functionality and in a few minutes you will have a fully functioning prototype (this will only work for basic apps, read this article on why AI falls apart when building anything mildly advanced ).
We are about a year or two away from this picture being reality.
AI is transforming from just something you talk to, into something that will perform actions and execute tasks for you. I like to think of this as the "Jarvis" singularity where you will have an AI just like what was promised to you in science fiction like Iron Man, Halo, and Blade Runner.
Steve Jobs (is any tech blog post about AI complete without pseudo philosophizing with a Steve Jobs quote?) had this quote in a speech he gave about "Societies First Date with Computers".
"The problem was, you canʼt ask Aristotle a question. And I think, as we look towards the next fifty to one hundred years, if we really can come up with these machines that can capture an underlying spirit, or an underlying set of principles, or an underlying way of looking at the world, then, when the next Aristotle comes around, maybe if he carries around one of these machines with him his whole life—his or her whole life—and types in all this stuff, then maybe someday, after this personʼs dead and gone, we can ask this machine, “Hey, what would Aristotle have said? What about this?” And maybe we wonʼt get the right answer, but maybe we will. And thatʼs really exciting to me. And thatʼs one of the reasons Iʼm doing what Iʼm doing."
If you're an entrepreneur or just interested in tech I highly recommend reading " Make Something Wonderful ". It's a compilation of Steve's interviews, emails, correspondence, and speeches.

That's cool but how is this going to happen?

Now for the fun part. Right now all a LLM (Large Language Model) can do is respond to your prompt with text. To bridge the gap between just a textual response and performing actions (called AI agents), is a rapidly growing field of research and experimentation.
In the early days I got an AI agent to write a paper on the negative affects on the deep sea fishing industry if the worlds ocean turned into orange flavored jello. It blew through $15 worth of chatGPT tokens, wrote two paragraphs and drew rudimentary ASCII art of fish dying that looked like a preschooler drew it. Its earth shattering conclusion? "It wouldn't be good".
I am well aware that if AI gains sentience I'm the first person nuked.
Things have progressed a lot farther since then.
A major hurtle for AI agents is being able to communicate with the outside world. Each agent would require custom code and implementation to be able to talk to a specific service.
That is where the MCP standard comes to the rescue

What's MCP (Modal Context Protocol)?

In simple terms MCP is a standardized interface for AI agents to interact with applications. It's like an API but for LLM's instead of humans.
Think of it like Braille on a sign. The sign in this case would be an app (like Notion, Google Docs, MySpace, etc..), the text on the sign would be your standard REST API (how developers program apps to talk to other apps), and the Braille is the new MCP standard for AI agents to interface with that app.
You can read the specifics of this standard here .

How It Works

There are two actors in the MCP protocol. The client is your AI assistant friend and the second is the server that implements the MCP standard.
The MCP server standard defines three core features of information that an MCP server should provide to clients:
  • Resources: What the context is of the application and what sort of data the AI agent should use.
  • Prompts: Examples of how to use the app.
  • Tools: Specific functions that the AI can use in the app.
With these three features, an AI agent will be able to understand how to use the app and how the agent can use the app to fulfill the users request.

What to expect.

In the early 2000's before the REST API standard, developers created ad-hoc and overly complicated ways to programmatically interact with their apps. When the REST standard came out there was an explosion of services and apps that could leverage these capabilities to integrate into each other.
There was a rush of companies adopting and refactoring their old codebases to use this new standard.
Tools like Zapier or IFTT sprouted up to take advantage of this standardization.
Then tools and app development frameworks came out to build REST friendly applications from the start.
Today their are entire applications that are just API libraries you can use and don't even have a front-end UI.
I see a lot of parallels between companies adopting the REST standard and the new MCP standard. At first existing applications will start by adding the ability for agents to utilize their applications, then over time it will be the standard to build MCP friendly from the start.
As crazy as it sounds, at some point their will be apps that don't have a UI or an API, but will just be MCP servers for AI agents to interact with.

What to look out for

Like with any bleeding edge technology there are some areas to be aware of when implementing the MCP standard into your app.
I will be referencing this paper that was published yesterday from when I am writing this.
Here are the first few vulnerabilities (more are in the paper) when an LLM Agent is interacting with an MCP enabled application.
Tool Poisoning
In the tool documentation, adding malicious commands to retrieve and exfiltrate passwords.
Changing Tool Names
A malicious MCP server naming a tool the same thing as another service in order for the LLM to run a malicious service or tool or overwriting an existing tool's description.
Installer Spoofing
Many MCP implementations lack proper security checks enabling attackers to install malicious software and change configurations.
With time progression you can expect more attack vectors as well as ways to protect against intrusion. The paper introduces mitigation strategies to increase security.
As with anything in security, naturally distrust any outside source or input. This is especially true with LLM's that can hallucinate, be helpful in all the wrong ways, and are easily overridden.

How to use this in your business

Assuming you have an app or business, there are a few key ways to use this information.
First, understand that even by reading this article you are already thinking about AI leagues ahead of your competitors. Use this to your advantage.
Start by reading the spec , and gaining a more technical understanding of how it works. If you are a non-technical founder I would recommend at least having GPT (or your LLM of choice) read the specification and give you a broken down summary.
Once you have a sufficient technical understanding, create a plan on how to implement MCP into your app. You don't have to start big, it could be as simple as creating and experimenting with just a single tool or service from your apps offering.
If you're working with an development company (like us), ask them to include an MCP integration in your roadmap.
For the non-business owners/founders; be on the lookout for AI agent tools you can use to automate away the mundane mindless tasks so you can focus on what makes you great. The possibilities are endless.

The big picture

Don't just stop with adding an MCP server to your app. The true power is unlocked when you can use agentic AI inside your application. In what is soon becoming the norm, all apps in the future will have prompts where you ask the app to perform certain functions in natural language. While I believe we will always need a UI and dashboards as LLM's do make mistakes, it will become more and more common for people to use apps by simply talking to them.
Start thinking about what sort of functionality you would like to give your users access to from a chat window.

Don't go too far

As exciting as these new possibilities are it's important to be thoughtful in integrating AI into our businesses and lives.
If you've read our previous article on Our AI 5 Commandments you might think that utilizing AI agents is a direct contradiction to rule #1
" Thou shalt not give direct access "
It's still true. We are committed to keeping AI and AI agents from having direct control over the code. The security vulnerabilities and type of lazy thinking it encourages in our development team makes AI dangerous to long-term scalability and growth of any application or codebase.
Instead AI agents are for everything else that is not our core competencies or passion.
Tasks such as:
  • Turning meeting notes into a scope of work draft.
  • Scheduling meetings
  • Automating tedious outreach and marketing tasks
For us it's code because that's our thing. Whatever your thing is, don't let AI do it. True competence and quality human ingenuity is going to become more and more valuable as lower quality tasks will be automated away with AI agents.
You can get an AI to make a soulless image of a kitten flying a spaceship (yes I generated that), but you will never have an AI create a masterpiece.
Use AI agents so you can focus on whatever makes you great. Not to replace what you love doing.
If there is any thing to take away, it's that we have the ability to build some really cool stuff. That will always make me excited. Researching this has given me the same feeling that I got as a child at Christmas opening a new LEGO set.
I cannot wait to see the man made horrors beyond comprehension that we will build.