MCP: When Chat Goes Beyond Just Text (and What We Did with It at PairProgramming)

There’s a quiet shift in how people search for and hire services. Fewer and fewer start on Google, and more and more begin by asking an assistant. "I need someone to build me an insurance comparator," "How much does it cost to digitize my clinic’s appointment system?", "I want an MVP to validate an idea." The conversation has become the place where decisions are made. And if your studio isn’t part of that conversation, you simply don’t exist in that moment.

That was the starting point for what we built at PairProgramming with MCP. We didn’t approach it as a technical experiment to tweet about. We approached it as a business question: if a potential client is chatting with an AI about their project, how do we get in there, be useful, and not interrupt?

What is MCP, without the boring part

MCP —Model Context Protocol— is an open standard that allows AI assistants to talk to real applications. Think of it as the universal plug between a model and the real world: instead of each assistant inventing its own way to connect to every service, they all speak the same language. It emerged in late 2024 and, in just over a year, became the de facto way the industry connects models to tools.

At first, MCP was an internal thing: it gave assistants superpowers, but everything ended up being resolved in text. You asked, the AI used a tool in the background, and it answered with words.

What changed the game in 2026 was MCP Apps: the first official extension of the protocol. The novelty isn’t minor. Now an application can return a real interface—a form, a dashboard, a multi-step flow—and that interface is rendered right inside the conversation. Clients like Claude, ChatGPT, Goose, and VS Code already support it. It’s the difference between an assistant telling you something and showing you something you can interact with without going anywhere.

And things get even more serious: on July 28 of this year, the biggest revision of the protocol since its inception was published. A sessionless core that scales on common HTTP infrastructure, a formal extension framework, authorization aligned with industry standards. Translated: MCP stopped being a toy for early adopters and is becoming the infrastructure on which you can build seriously, for the long term. That’s exactly the moment you want to be inside, not watching from the outside.

What we built in the PairProgramming app

We set up our own MCP server for PairProgramming. In plain terms: we opened a door for AI assistants to interact with us directly, in the middle of a chat, without sending the user to fill out a form in another tab or wait for an email.

We gave it three capabilities, designed from the perspective of the person on the other side, not the architecture:

Scope a project on the spot. Someone describes what they need—"I want a site with reservations and payments"—and instead of a generic response, the assistant can translate that idea into concrete PairProgramming services, with a sense of where to start. The conversation stops being abstract and becomes a proposal.

Leave a lead without friction. If the person wants to move forward, we don’t send them hunting for a "Contact" button buried on the site. The interest is recorded right there, at the exact point it appeared, when it’s hottest.

Show an embedded contact form. And this is where MCP Apps shines: instead of spitting out text asking them to copy their email, a real form appears inside the chat, neat, to fill out in two seconds.

The detail I like most about all this—and it’s easy to overlook—is that every action triggered by that interface goes through the same control as any other. There’s no backdoor. A click on the form has the same level of consent and audit as any assistant action. For something that handles real people’s contact data, that’s not a footnote: it’s the difference between being able to sleep soundly or not.

Why this matters (especially if you’re small)

There’s an easy take that says "well, it’s just a contact form with a twist." No. What changes is where the conversation with your brand happens.

For twenty years, the game was to bring people to your territory: your website, your landing page, your funnel. MCP flips that. Now you can go where the person already is—inside the assistant they chose to use—and be useful there, on their terms. For a bootstrapped studio like ours, which doesn’t have a marketing budget to fight for attention with sheer volume, being present in the right channel at the right time is worth much more than shouting louder.

And there’s something almost poetic for a studio called PairProgramming: for years, we helped others build on new technologies. With this, we get to eat our own dog food. We implemented the standard while it was still being defined, understood its limits hands-on, and now we know what we’re talking about when a client asks "does this MCP thing work for me?" The answer is no longer theoretical.

Where it’s headed

The protocol is at the exact moment when things get interesting: mature enough to build on with confidence, early enough that almost no one is taking advantage of it yet. That window won’t stay open forever.

What we built at PairProgramming is a deliberate first step. The conversation is the new interface, and we’d rather learn to inhabit it now—with our own real, production case—than arrive late when it’s already obvious to everyone.

If you’re thinking about how your product or business can live inside that conversation instead of competing against it, it’s literally the kind of problem we like to solve—from custom AI agents to the entire product. Get in touch.

---

By Esteban Aleart, Founder & Lead Engineer at PairProgramming.