Are your users already searching for your MCP server?
Developers at companies of all shapes and sizes are already wiring their apps into AI agents.
Not next year. Not “someday.” Right now.
They’re plugging their tools into ChatGPT, Claude, and the new wave of IDEs with first-class MCP support. They’re building workflows and prototypes they would’ve never attempted even twelve months ago.
And when your app isn’t part of that ecosystem, they’ll still try to connect it in.
The problem?
When you let an LLM freestyle API calls, the results are inconsistent at best, and often bad enough to make them stop and look for another tool that gives them better results. That's if it works at all.
Let’s talk about what’s happening, what your users are actually doing, and why this matters more than most teams realize.
The scale is bigger than it looks
The truth is that we’ve never really seen user adoption of a new consumer technology at this scale, let alone this fast.
Sure, not every workflow runs through the biggest clients, like ChatGPT and Claude. Some teams are building custom agents. Some lean on local models. Some do both.
But with over 800 million ChatGPT users, the reality is straightforward: the vast majority of your users are going to engage with AI in environments that support MCP.
This isn’t niche anymore. You can't get any more mainstream than 800 million users.
And over the last year MCP has quietly become a major piece of glue holding all of this together—while still giving users the creative freedom they expect from their favorite AI tools.
What your users are actually doing right now
I mostly work in the space of large consumer apps, so I have a bit of an inside scoop on the broader patterns for how these tools are (and aren’t) being adopted.
And it's a different adoption path from anything I’ve seen before.
People aren’t filing feature requests asking for “MCP support.”
They’re asking ChatGPT if it can do something with your app… and hoping it answers correctly.
They’re searching GitHub for “your-app MCP server” to see if someone else already built one, even if it's not you… and hoping it represents your brand well 🤞
They’re browsing repos for high-quality servers the community trusts… expecting to find yours.
They’re having a moment to imagine:
“It’d be great if I could use [YOUR APP] with ChatGPT.”
“I wish I could wire [YOUR APP] into my agent workflow.”
When that moment happens, you want to already be there. Instantly usable, friction-free, and ready.
Why MCP matters: friction kills adoption
People want to stay where they already are. Even early adopters like me would rather keep the majority of our focus in the places we already know. We've already chosen my favorite tools and configured my environment to my liking.
When someone is deep inside ChatGPT, Claude, Cursor, VSCode, or their custom agent, asking them to switch tools to work in your app breaks their flow.
Switching introduces friction, and friction kills creativity.
Instead of building a chatbot into your app, MCP makes it easy for anyone to now work with your app from inside their chatbot of choice. And, with the exception of niche use cases, their chatbot can serve them thousands of times better than yours can (Stop Making LLM Wrappers). Even in the realm of niche use cases, professionals are using MCP to power the agent-to-service communication.
When you ship a well-designed MCP server, it puts you right where they want you, right in the moment they need you. No tabs, no context switching, no extra scaffolding.
A great MCP server is not “just wrapping an API”
If all you do is expose endpoints, the experience will be mediocre for users and confusing for the LLM. It will often choose the wrong endpoint, or get the data structures wrong.
The end result? A messy experience for your users, full of failed experiences that waste tokens and erode user trust.
A well-crafted MCP server is purpose-built:
- Designed for LLMs and agents, not humans
- Structured so the model understands your data and capabilities
- Sculpted for natural-language workflows
- Clear, predictable, and optimized for agent reasoning
- Polished with tool descriptions, user-centric affordances, and the right abstractions
I’ll be sharing more about this design process in coming emails, including more examples of what can go wrong and how to avoid the most common mistakes.
But done well, it works in ChatGPT, Claude, Cursor, VSCode, custom agents, and anything else that supports MCP.
You build once. It Just Works™️ everywhere.
That’s the Epic way. Build with standards.
This is about creative empowerment
Remember, your users already have ideas they want to try, but they stop because they have to figure out a bunch of things before they can even start. A great MCP server removes the friction that stops them, by
- Unlocking workflows they wouldn’t have attempted before
- Letting them stitch your app into their existing agent setup
- Making your product feel like a native part of their creative environment
- Turning your service into a capability they can mix, match, and extend
When your MCP server is ready, your app can become a seamless part of their creative toolkit, automatically.
The path forward
MCP gives you the tools to build these integrations the right way:
- Designed for agents
- Designed for natural language
- Designed to delight
- Designed with the craftsperson mindset your users will feel immediately
If you want to build servers your users will love (and that AI agents can use effectively) there’s a right way to do it.
Learn how in Epic MCP
If you want to build MCP servers that actually work the way users expect, now is a fantastic time to enroll in Epic MCP.
You'll learn…
- How to create a server that leverages the most useful features of MCP
- The vocabulary of MCP to navigate the new ecosystem
- How to add UI to get the best of natural language and familiar graphical interfaces
- How to authenticate users and access user-specific data
- The very latest features and techniques for building effective, user-friendly and LLM-friendly servers
All at your own pace. You’ll be equipped with the knowledge you need to go back to work and connect your services with the agents your users are using.
Despite the massive adoption, we're still in the early days. Now is the best time to get your app ready for the way users are building today.