This Friday we’re talking about the hottest GTM channel–search in LLMs. Some call it AEO, some call it GEO, but call it whatever you want figuring out how to get Claude, ChatGPT and Gemini to recommend your product or service is the #1 fastest growing growth channel.

Stan Chang, head of Product at Gumshoe.ai is going to break down this whole channel and show us tactical tips for how to win. Big surprise for me is that while SEO takes a long time to change–AEO can be changed in a month or two. Learn why this Friday!

Onto our discussion of MCPs, the interfaces AIs use to talk to each other (yes, that’s a simplification–but only slightly).

For 70 years, humans have had to learn to talk to machines.

Every generation, we get a little better at it. The machines get a little more forgiving. But the fundamental arrangement never changed: you wanted something from the computer, and you had to learn its language to get it.

Until now.

In 1955, you didn't use a computer. You submitted a request to someone who did.

Specialists wrote programs on punchcards — one instruction per card, a stack of them rubber-banded together — and handed it to an operator. The operator fed it into a machine the size of a living room. You came back the next morning for your results.

If there was a typo on card 47, you got nothing. Start over.

Computing was a priesthood. Access required credentials, institutional affiliation, and physical proximity to a very expensive room. IBM's customers weren't people. They were governments and corporations with seven-figure budgets.

This was not considered a problem. It was simply how computers worked.

The minicomputer changed the room. The PC changed the building.

By the early 1980s, computing had moved from specialized rooms to office desks. But it still spoke its own language. MS-DOS. The command line. C:\> staring at you, waiting.

If you didn't know the commands, you got nothing.

Then came the GUI. The mouse. The icon. Windows.

Suddenly, anyone who could point a finger could use a computer. You didn't need to memorize syntax. You clicked on the thing that looked like what you wanted.

A hundred million people who couldn't type DIR /W became computer users overnight. New industries — desktop publishing, spreadsheet-driven finance, word processing — appeared in years, not decades.

The interface changed. The world changed with it.

The PC conquered the desk. Mobile conquered everywhere else.

In 2007, most of humanity had never owned a computer. By 2015, three billion people carried one in their pocket.

No mouse. No keyboard. A screen you touched with your thumb.

Nobody needed a high school class to learn how to use their mobile phone–it was intuitive.

The shift wasn't just hardware. Apps replaced websites. Location replaced the browser. Services that couldn't exist on a desktop — Instagram, Uber, WhatsApp — were built mobile-native and became some of the most valuable companies in history.

The interface changed again. Billions of new users. Entirely new categories.

Every one of these shifts made computing more accessible.

None of them solved the fundamental problem: you still had to learn how to use the software.

Commands became menus. Menus became touchscreens. But somewhere underneath every piece of software was a model of the world that you, the human, had to internalize before you could get anything done.

For consumers, video game companies learned to make things easy to figure out. Churn was real and fast. But the enterprise was insulated from this reality: some higher up bought the software, and your job was to learn to use it.

No kid ever woke up one day and said “Mom, when I grow up I want to become an expert on clicking through menus in a health care billing software,” yet tens of thousands of people do that every day. And you can replace health care billing with nearly any enterprise software package.

Not only is there friction in learning a new tool, to the point that people stick with old, inferior tools simply to avoid retraining, but also the gap between what software can do and what most people actually get out of it is enormous. It always has been.

Until now.

MCP — Model Context Protocol — changes the question.

Not "can you use this software?" but "what do you want to accomplish?"

For the first time in the history of computing, the machine learns your language. Not the other way around.

This isn't a chatbot. A chatbot answers questions. MCP lets AI agents take action.

MCP describes a type of server, a Model Context Protocol server, that enables AIs to communicate. It tells the AI what it can do, and does it.

You can use MCPs to read your CRM, research a prospect, send an email, update a record, pulls a report or even writes a blog post or a slide deck–without a human navigating a single menu, all from an AI tool like Claude.

And critically: it isn't automation. Automation is scripting clicks. You still had to build the script. You still had to maintain it. You still had to know what you were automating.

MCP is intent → outcome. With nothing in between.

The AI can ask you questions to get jobs done that aren’t necessarily supported. “I can find that contact, but I’m not able to update it. Do you want me to copy it and create a new one with your changes?”

Want a report? Claude can use the MCP to build you a report in any format you want. No “reporting” UI to learn, no analyst to ask for help.

On a societal level, this change is huge. Where mobile added three billion users by making computing portable, MCP removes the need to learn anything for everyone. Just talk to the computer, it will do the complex task for you.

But the bigger shift is in business software. There is no “retraining” if everyone is logging into Claude and telling it to access this or that MCP provider to do a task. If a better vendor comes along, Claude might just know–you might not even have to tell it to switch what it uses.

I use Exa today because Claude told me it was better at a specific task than Perplexity. Ouch! (See also: our event on getting LLMs to recommend your product this Friday!)

A vendor asked me if we’d consolidate our contact enrichment with them. Why would we? We have 10+ providers, all good at slightly different things, and a complex workflow to reach out to each via MCP or API?

Why would anyone ever give that up, and go all-in with one vendor?

The companies built MCP-native — designed from the ground up for an intent-driven world — will look as strange and powerful to today's SaaS companies as Instagram looked to MySpace.

I was shocked at how slow the incumbents are right now. Skyp launched an MCP–and we’re one of the very very few in the email space.

Sales and marketing are the most interface-heavy functions in any company.

The average sales rep touches six tools in a day. CRM. Sequencer. Enrichment. LinkedIn. Email. Dialer. Each one requires context-switching, manual entry, and training time. Each one is a small tax on the hour.

Any rep can use an MCP. Open up Claude (or Cursor), and ask it what you want to do.

Now you can use Skyp to find prospects and send them emails–without learning another tool, but also without outsourcing the task entirely to AI. It’s humans using software on their terms.

The companies that win today won't be the ones with the best UI.

They'll be the ones that make the UI irrelevant.

See the MCP in action at skyp.ai/demo.