For the Love of God, Build an MCP
I built a Claude MCP connector for Flyletter, my AI newsletter writing tool, as a marketing stunt.
Now I'm building the entire product around it.
What changed was how I work. I run almost everything in Claude now, from my LinkedIn outreach in HeyReach (I've never seen its interface) to my newsletter in Beehiiv. It’s undeniably more efficient.
If I can't plug a tool into Claude and use it in a chat, I won’t use it. It’s way too cumbersome.
So I posted this:
It got over 11K impressions. 112 people liked it and 55 replied (not to mention a few reshares), so I'm clearly not the only one.
That can be a scary thing when you run a software company. It finally dawned on me that my customers won't use Flyletter without a good one, either.
If it's true for my product, it's true for yours. So let’s take a look at where software is headed, what I'm doing about it, and how to build your first MCP before your users move on without you.
Your Next Customer Won't Open Your App
Soon people will access software and the internet entirely through AI agents.
Which sounds bigger than it is. When I say agent, what I mean is a Claude chat connected to the software you already use, like your CRM. You tell Claude what you want, and it does the work inside those tools.
An MCP connector is a plug-in that lets you use tools within Claude. Without one, you have to leave the chat and use the software yourself, or have Claude take over your web browser and click through it, which is slow and finicky.
I made the full case for where software is headed in SaaS Isn't Dying, It's Just Losing Its Head, including why people won’t even look at your interface anymore.
How can I be so sure? I already run most of my tools this way, and I get more out of them with less work. For example:
HeyReach
HeyReach automates LinkedIn outreach. I connected Claude to HeyReach's MCP, and it sets up my campaigns, manages the messaging, and runs the analysis and reports, all in one chat.
Earlier this month, Claude found a campaign that wasn't sending any messages after people accepted my connection request. It caught the problem faster than I would have, suggested a fix, and implemented it without me ever touching the app.
Claude's also running A/B tests on my offer right now. Because an agent is analyzing my campaign data every day, I improve it constantly instead of whenever I remember to log in.
Beehiiv
When I re-launched my newsletter (this one), I had to get each draft into Beehiiv. Their API was gated behind a higher-tier plan, so I had two options: upload everything by hand, or have Claude take over my browser and click through it. Both were painful.
Then I found Beehiiv's MCP. Now it takes seconds.
It's also more flexible than any upload button. I can tell Claude to "use this template" or "match the formatting from last week's post," and it does. The same chat rebuilt my welcome email, launched a referral automation, and sent follow-ups to subscribers who never opened, all without me opening Beehiiv.
Once you work this way, a tool without a connector means leaving the chat, copying, pasting, and coming back. So it's the first one you cancel.
I Stopped Using My Own Product
I'd built Flyletter's MCP connector as a marketing stunt, something to announce and ride the AI wave. The app was the real product, and the connector's output wasn't as good.
When I was testing and iterating, I used the app. But when a real deadline hit, I stayed in Claude and used the connector. It was way easier to use, even though the app gave me better output.
After a while, I realized that I'd stopped using my own product. If I did that with something I built, my customers definitely would, too.
If Flyletter was going to survive, the connector had to become the core product.
So I'm rearchitecting Flyletter around the connector. The connector becomes the core feature, and I'm changing the way I communicate the product to match.
In the new version, Flyletter is the brain behind the writing. It builds a brand's voice profile and brand context, manages its content runway, and pulls in a live feed of its writing samples, so the agents always work from the latest examples. Claude gets read and write access to all of it.
Then Flyletter's agents run right inside the chat. One drafts a newsletter in your brand voice, another turns a finished issue into social posts, and another saves future topic ideas the moment they come up. All without leaving your Claude workflow.
How to Build Your First MCP
If you build software, don't get left behind. A product without an MCP connector doesn't fit the new way people interact with tools, so they won't use it. Build yours in three steps.
1. Start with read tools. These let Claude pull your users' data out of your product and into their chat. They come first because reading can't break anything, and it pays off right away. That's how Claude caught my stalled HeyReach campaign, by analyzing the campaign data every day.
2. Then add write tools. These let Claude create and update things in your product on your users' behalf. That's when Claude can do the actual work for them. When I tell Claude to "match the formatting from last week's post," a write tool is what builds the draft in Beehiiv.
3. Make it as good as your app. When users are busy, they'll reach for a connector even if it's worse because it's way more efficient. I did it with my own product. That makes the connector the thing your users judge you on, so give it the same care you give your app.
I walk through my full build, prompts included, in How I Built a Claude MCP Connector That's Actually Good.
And if you're the one buying software, ask whether it has an MCP before you pay for it.
Bottom Line
If I can't plug your tool into my Claude workflow, I won't use it. I learned that rule applies to my own product, and it's about to apply to yours.
Over the next few weeks, I'll show you how all these tools connect, starting with my newsletter workflow and then my full outreach pipeline.
What's one tool you wish had an MCP?
Writer, founder, and AI builder. Currently building Flyletter, previously sold Selling Signals (100k subscribers). 1 sold, 2 failed, 1 active. This post started as an issue of my weekly newsletter.