What an MCP app is (and why this feels like 1995 again)

Something like the 1995 website moment is happening again. A website made your business findable. An MCP app goes a step further: it lets the AI assistants your customers already use take real actions with your business, like booking an appointment or placing an order.

The 1995 parallel

In 1995, putting your business on the web felt optional, eccentric even. Why would a bakery need a home page? But the owners who did it anyway became the easiest businesses to find, and being easy to find turned out to be most of the game. The ones who waited too long disappeared from the place customers had started looking.

The same kind of shift is underway now. Customers are beginning to hand errands to AI assistants: “find me a plumber who can come Thursday, and book it.” Your website can tell that assistant about you, but it can’t take the booking. That’s the gap an MCP app fills.

So what exactly is an MCP app?

An MCP app is a small, secure piece of software that lets AI assistants take real actions with your business, within rules and limits you set. Check availability against your real calendar, build a quote from your pricing rules, place an order, answer a question from your actual policies. It connects through MCP, a vendor-neutral open standard governed by the Agentic AI Foundation under the Linux Foundation, so it isn’t tied to any one AI company. The full story of the standard itself is in What is MCP?

Your website was written for human eyes. An MCP app is the same business, made readable and usable by the software that now shows up on customers’ behalf.

What an MCP app can do today

None of this is speculative. Depending on your business, an MCP app can already let an assistant:

• Book appointments, reservations, and service calls against your real calendar, not a guess.
• Quote a job by collecting the details and pricing it with your rules.
• Order through your existing systems, with payment handled by providers you already trust.
• Answer questions from your own information: hours, availability, policies, what’s in stock.

The infrastructure is further along than most people assume. Stripe, Shopify, and PayPal all run official MCP servers, and more than 25 official remote MCP servers exist from companies like GitHub, Notion, and HubSpot. Every major assistant (ChatGPT, Gemini, Microsoft Copilot, Claude) supports MCP, and by Anthropic’s count in December 2025 the standard sees 97M+ monthly SDK downloads across more than 10,000 active public servers. When the payment companies are serving AI agents directly, the “is this real” question is settled.

A naming detail: MCP Apps, capital A

One naming note worth knowing. “MCP Apps” with a capital A is the official extension of the standard for interactive experiences, production-ready since January 26, 2026. It lets a business present a real interface inside ChatGPT, Claude, and other assistants: a booking calendar the customer can tap, a product picker, a quote summary they can confirm. The conversation and a small storefront, together, in the place the customer already is. The early web went from text pages to storefronts, and the same progression is happening here, compressed into a couple of years.

The early-mover math, again

An assistant completes the task with whoever it can connect to. It doesn’t weigh charm or reputation the way a browsing human would. If your competitor is bookable inside an assistant and you aren’t, you don’t lose the comparison; you’re not in it. That’s roughly what happened to the businesses that skipped the web in the late ’90s, just slower.

“Move early” doesn’t mean “do everything now,” though. One workflow done well, with clear limits, puts you ahead of competitors who’ve done nothing. For most businesses that’s the bookings that interrupt the day, or the quotes that pile up overnight.

What you actually get out of it

Every booking an assistant completes is a phone call that didn’t interrupt your team mid-job. Every routine quote handled overnight is an hour back in the morning. We started Andromesh Labs on the theory that this is what software is for: taking over the work that never needed a person, so the people can do the part that does.