The Hidden Revolution in Our Digital Plumbing
Unless you’ve been keeping a keen eye on your AI news, you might have overlooked the announcement and rapid adoption of Model Context Protocol (MCP). Infrastructure changes tend to happen quietly beneath the surface while we're distracted by flashier developments. But just as APIs silently revolutionized application development in the 2000s, the Model Context Protocol is laying the groundwork for a profound shift in how we'll interact with the internet for decades to come.
To explain MCP, I’ll need to back up a bit and explain the basic underpinnings of the internet. If you already understand APIs and client-server interactions, you can skip this part.
APIs explained
The internet is really just a bunch of computers talking to each other through an amazingly simple but powerful network protocol called TCP/IP. When two computers want to talk, they look each other up in an address book called DNS. When you buy a domain name, you’re really just buying a listing in the address book and pointing your domain name to a specific computer. Usually you put your website on that computer, which then serves it up to anyone who asks for it. That’s why it’s called a server.
When you type a url into a browser, you are reaching out to the server where that website lives and asking for some data (the html to present the website). You basically say, “I want to see this page—here’s my address.” And then the server starts spewing out little packets of information with your to: address attached to them. Those little packets fly out to the network taking many paths, eventually landing and getting reassembled by your computer.
Almost everything on the web works this way. Facetime calls, Netflix movies, Zoom meetings, anything on your browser, etc. Now you know how the internet works. For a mind-blowing deep dive on this, check out this video.
As websites, apps, and web-apps started getting more sophisticated, they began to rely on more than their own server to perform their functionality. For example, let’s say you have a travel site that allows people to book vacations. Your application might want to access more specialized servers to get plane ticket pricing, hotel availability, weather in faraway destinations, etc. It would be silly for you to build all that when somebody else has already done it.
So, modern apps usually call APIs (Application Programming Interfaces) on other servers to get all the data they need. Your travel app might use Amadeus for airfare ticket prices, Booking.com for hotel booking, and OpenWeather for weather. Some APIs are free, but most of them charge a little money for every transaction. Your travel app might even publish its own API to allow other applications to book travel on your application.
APIs have become a major backbone for functionality on the internet. Almost every non-trivial app and web app uses an API somewhere in its code. Now you know what APIs are and why they are useful.
Agents, tools, and MCP
Now, let’s talk briefly about Agents. You can read my deep dive here, but to avoid belaboring my explainer any further, I’ll summarize. Agents are AI powered automations with some sort of autonomous capabilities. You give an agent a task, and it uses an LLM to decide how to accomplish it given a certain number of tools at its disposal. There’s a lot of hype and no official definition of what an agent is, so I wouldn’t blame you for being totally confused about them. A lot of people call any AI automation an agent, but most tech people agree with Anthropic’s concise definition.
You could tell your agent: “Send an email to my brother-in-law and politely tell him I won’t be going to his 40th birthday party in Tijuana.” To accomplish this task, your agent would need a few tools: your address book, your calendar, and access to your email account. If your agent knew it had these tools it could easily accomplish this task. If you asked ChatGPT to do this (it doesn’t have those tools) it would be flummoxed.
Okay. We’re finally getting to the juicy part. Once you start building agents, you really want to hook them up to useful tools. And lo and behold, we’ve spent the last 20 years building APIs just for this purpose. Well, we built them for human developers to use, but they are just as useful for AI agents.
I could build an agent that looks up flights for me. But. I would have to describe the API to the agent so it knows how to “talk” to that server. It’s doable, but it’s a bit of work. It would be a lot nicer if the server could just tell the AI how to use it, sort of like an instruction manual written expressly for an AI agent.
And that’s what MCP is. It’s a protocol that servers can use to tell an AI how to access their functionality. Take that in for a second. All the web infrastructure we’ve built over the last two decades made available to AI agents.
The thing is that agents need a consistent way to get this information. So in late 2024, Anthropic open-sourced the Model Context Protocol (MCP), creating a new standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments. But this technical announcement obscures a much bigger story about how the internet itself is evolving.
From Passive Systems to Active Agents
Here's the thing about traditional internet infrastructure: it's dumb. Not "bad" dumb, but passive dumb. Your browser doesn't understand what you're looking for. Your email client doesn't know what messages matter. Your documents don't recognize what information you need from them.
MCP fundamentally changes this equation by enabling AI systems to act with agency across internet resources.
MCP is often described as like a USB-C port but for AI agents: it offers a uniform method for connecting AI systems to various tools and data sources. The wow insight: your AI assistant can now independently discover what resources exist, determine which ones might help you, and take action to access them—all without explicit instructions for each step.
This isn't just a marginal improvement. It's a categorical change in how internet infrastructure operates. But it hasn’t quite happened yet.
Why This Changes Everything
Think about how you currently navigate the web. You need to:
- Know what resources exist
- Remember how to access them
- Manually navigate to each one
- Extract the relevant information
- Synthesize it to solve your problem
The cognitive load is entirely on you. It's like having a library where you not only need to know which books hold your answers, but you also need to manually retrieve them, flip to the right pages, and connect the dots between different sources.
MCP reshapes this paradigm by allowing AI systems to securely and dynamically interact with data sources in real time, replacing multiple custom integrations with a single, standardized connection method. This means your digital assistants can independently:
- Discover what resources are available to them
- Determine which are relevant to your needs
- Access those resources in real-time
- Process the information contextually
- Take action based on what they learn
For users, the infrastructure becomes invisible. You express an intent, and the system handles the messy details of actually finding and using the right digital resources.
The Real-World Impact
This isn't theoretical. Early implementations already show the potential:
Using an MCP server for tools like Puppeteer (a web scraping tool) allows AI to automate complex web interactions by allowing it to interact with web pages. This means AI can not only search the web but actually navigate it like a user would, clicking buttons, filling forms, and extracting information.
If your codebase is on GitHub, an MCP server can fetch specific files without requiring a complete clone of the repository. The Github MCP server can also create branches and help with issue management. Software development becomes a conversation with systems that understand code context and repository structure.
These early examples only scratch the surface. Right now the main users of MCP are developers, using it mostly to enhance programming tasks. But as the protocol matures, we'll see AI systems capable of orchestrating complex workflows across dozens of digital resources without explicit programming for each step.
There is a certain “looseness” to this that’s hard for me to get my head around, but it aligns with the general paradigm shift we’re seeing with AI development. Programming is becoming less about exhaustive explicit instructions and more about expressing intent and letting the AI work out the details.
In an MCP world, I might tell my agent, “I want new running shoes.” The agent could then “understand” that intent, look at the resources available to it, help me refine my intent, and then go present and purchase some shoes for me. Online shopping could be radically transformed. It doesn’t take a think tank to see where this is going.
The Profound Shift Ahead
Here's why this matters beyond the technical implementation: MCP represents the first step toward giving AI systems true agency within our digital infrastructure.
Today's AI systems are like well-trained but confined librarians, limited to a specific collection of books. MCP gives them the ability to venture beyond their designated shelves, discover new sources of information, and utilize tools they weren't explicitly programmed to use.
This shift toward AI agency across the internet will fundamentally change how we interact with digital systems. The endless frustration of "which app has that file?" or "how do I connect these two services?" begins to disappear when your AI assistant can independently discover and leverage resources across your digital ecosystem.
It also has profound implications for digital middle-men. Remember the travel website we were talking about earlier? Why would someone need that if my agent can go directly to the source to find the information it needs? Why go to Zappos when my agent can find all the shoes from all the manufactures directly? What will happen to banking or brokerages if the interface to transactions is your agent?
Most of the tasks we perform online today are really just user interfaces (form fields, pull-down menus, and buttons) on top of various APIs. A user interface is basically friction between us and what we want. We’re so used to the status quo that we don’t even think about it. But what does the world look like when the only interface we have is between us and our agent?
The prospect of a browser-less internet can be hard to get your head around if you've been working in digital for a while. SEO becomes moot. Whole applications that are currently written in a deterministic way could be replaced with ad-hoc agentic workflows (e.g. CRM, ERP, EHR, etc.) Why would we need dashboards or BI tools? In fact, you could view the whole concept of "browsing" as essentially friction. I generally avoid treading in futurism, and based on the state of things today, we're a ways out from any big disruptions, but this is one area where I think we need to think ahead in order to be prepared.
What's Next?
We're still in the early days. MCP is a baby standard and everyone is figuring out how to use it and discovering its strengths and weaknesses. For example, MCP has some pretty large security holes that will need to get ironed out. But the real transformation will come as this standard (or one like it) proliferates and becomes embedded in everyday digital infrastructure.
The internet is about to get a lot smarter—not because the AI models themselves are improving (though they are), but because the infrastructure connecting them to the world's information is fundamentally evolving.
Those who understand this shift early will be positioned to build the next generation of transformative digital experiences. The rest will wonder why their meticulously designed applications suddenly feel outdated in a world where digital resources respond intelligently to human intent rather than explicit commands.
Keep your eyes on MCP—the internet plumbing revolution has begun. Here at Machine & Partners, we help companies navigate and find opportunity in this rapidly changing world. Give us a shout if you’d like us to think through how your business might be affected.
