Could you explain MCP to my friend?

Apr 11, 2025

Yes, it's definitely my friend who doesn't understand the hype and the acronyms.

The next step towards Agentic AI is transferrable memory - not just the expanded memory that OpenAI is rolling out in Mid-April, but across models and applications. What sort of memory do current models support, and why is MCP fundamentally different?

Wait, What’s MCP?

“MCP (Model Context Protocol) is a protocol. It’s an open standard for how MCP clients/servers can interact.”

Translation:
MCP is a shared language that helps different AI tools talk to each other—especially when it comes to who you are and what you care about.

Imagine your CRM, email assistant, and AI co-pilot could all tap into the same profile of you—your tone, your goals, your preferences—and respond accordingly. MCP is the plumbing that could make that possible.

The creators of MCP like to think of it as the USB-C port of the AI world - a universal connector to all things.

Why Is It Taking Off?

“Adoption of MCP is growing fast, because it’s relatively straightforward to understand and build with but has enough specificity to be useful.”

Translation:
It’s gaining traction because it’s simple, useful, and powerful. Developers can build with it quickly—and it solves a real problem: persistent context.

No more repeating yourself across tools. AI can finally remember your priorities and adjust accordingly.

The Fine Print (aka The Current Problems)

Dharmesh calls out four big challenges:

Authentication – Who’s allowed to load or use your context?
- Business risk: Could a competitor hijack your context data?
Trust – How do you know an MCP server is secure or legit?
- Reality check: Just because someone offers “MCP support” doesn’t mean they’re trustworthy.
Provisioning – MCP is still DIY.
- Today: You need technical chops to set it up. No one-click install (yet).
Security – New connections = new risks.
- If AI knows your company secrets, you better make sure it doesn’t leak them.

So Where Are We Today?

We’re still in the builder phase. If you’re not technical, this isn’t plug-and-play yet. But the groundwork is being laid—and fast. But you should care because the promise of MCP is a large pool of "x" clients being able to connect to a large pool of "y" servers in a language that translates instantly.

Imagine a future where:

ChatGPT remembers your brand voice
Notion updates your roadmap based on voice notes
Your CRM tunes outreach based on your real goals
Gamma can apply brand standard changes to saved themes across 100+ presentations
NotebookLM doesn't need 15 new source links to plan your next trip because it's connected to your American Airlines and Marriott accounts already

And they’re all pulling from the same context file.

This is personalized AI at scale—without having to re-train every tool. The "holy grail" of consumer applications seems to be the true, personalized digital assistant. If we want something that could replace human assistance, giving it a combination of human memory and ability to adapt that memory no matter the task would be critical… right?

Bottom Line

MCP isn’t just a nerdy side project. It’s a signal.

Yes, it’s early.
Yes, it’s messy.
But if it matures, MCP could unlock AI that’s not just smart—but actually knows you.

Want help preparing your business for what’s next?

This is the kind of shift where the early adopters don’t just get ahead—they leap ahead.

Let’s talk about how to future-proof your AI strategy.