The next app store is a conversation

While you’re trying to get someone to visit your website, your competitor might be getting installed inside the software they already use.

That’s why I’m paying attention to Claude’s plugin directory. The interesting part is where it puts your product: closer to the moment someone needs a job done, with a way to do that job inside the conversation.

At AppSumo, we also built Sumo. When we were growing it, we looked for WordPress plugins whose owners had stopped paying attention to them. The software might have been neglected, but the people using it were potential customers we wanted to reach.

We bought plugins, often for less than a few thousand dollars, and could estimate the exposure Sumo would get from a promotion inside their dashboards. We could weigh the price against the chance to reach those users before spending the money. Distribution was part of the buying decision from the beginning.

That’s the part I recognize here. While everyone is discussing what the new technology can do, I want to know who it lets me reach. You have to take advantage of what others aren’t seeing, and sometimes it’s right there in the announcement.

The announcement I would take seriously

On September 25, Anthropic opened a new submission portal for Claude plugins. Developers on paid Claude plans can submit a remote MCP connector or a plugin bundle, follow its review, and publish after approval. The portal also provides discovery and usage information, including listing views, search terms, and installs.¹

In plain English: there is a place to put software in front of people already using Claude to work, and a way to learn whether they find and install it.

A connector lets Claude work with another service. A skill supplies instructions for doing a particular job. A plugin can package those pieces together. You don’t have to rebuild your entire product to find out whether one useful part belongs there.¹

The detail I care about most is that this can go beyond returning a paragraph of text. MCP Apps can show an interactive interface, such as a chart or a map, inside the conversation.² The customer can ask for something, inspect the result, and keep working without making your dashboard their next destination.

That changes the question I would ask when planning a product. Alongside “How do we get customers to our website?” I’d ask: “Where could our product do useful work before they need to visit it?”

Your homepage might be the long way around

Imagine you sell software that helps an agency prepare reports for clients. Today, your pitch might ask someone to visit a landing page, start a trial, connect their accounts, learn the interface, and build a report.

Now imagine the agency owner is already working in Claude and asks which client campaigns need attention. With the right connection and permission, your product could supply the analysis right there. They would still need to discover and connect your product. But once connected, it could become part of a task they already start in Claude.

I would rather investigate that than spend another month debating the button on a landing page nobody visits. The button still matters. So does onboarding. But a founder can perfect both while avoiding the harder question of how anyone will encounter the product in the first place.

There’s a difference from the WordPress example that makes this worth watching. A dashboard promotion asks someone to stop what they’re doing and consider another product. A useful plugin could help them finish what they’re doing. Your product gets its chance in the middle of the work.

An integration badge won’t save you

There’s an easy way to waste this opportunity: expose every feature as a tool, put “works with AI” on your homepage, and count the integration as finished.

An agent still has to understand when your product is useful and how to use it. Then your product has to deliver. Anthropic’s engineering guidance recommends testing tools against realistic tasks, inspecting failures, and improving descriptions and responses. It also emphasizes choosing useful tools rather than simply wrapping every existing function.³

I’d count that work as part of distribution. Being available means little if the agent can’t tell when to use your product, or the customer gives up after the first attempt.

For that reporting business, I’d start with one job: identify which campaigns need attention and explain why. I’d want the answer to show the underlying numbers, say what period they cover, and make missing data obvious. Only after that works would I worry about reproducing the rest of the dashboard.

My bet is that a smaller competitor doing one important job reliably could earn a place in the customer’s routine while the larger company is still deciding how much of its product to bring over.

You’re still building on rented land

Getting listed doesn’t promise discovery, and getting installed doesn’t mean anyone will pay. The directory gives you a way to test demand, with rules you don’t control.

Anthropic reviews submissions and can require changes to keep them in its directories. Its policy explicitly includes ongoing review and future requirements.⁴

My first Facebook game reached more than a million users in less than a month. I take new platforms seriously because I’ve seen how quickly you can reach people through them. I also care about what you own after that first burst of attention. At AppSumo, we owned our email list, so we could make the next offer to an audience we could reach again.

I’d approach Claude with the same question: if this works, what are we building beyond a listing? Paying customers who want to keep using the product are a much better reason to invest than a place in a directory.

For a plugin, I’d want to know who actually gets value, who returns, and who becomes a customer. A user who chooses to create an account with your service is a different business outcome from an anonymous tool call. Respect the permissions involved, explain what’s paid, and make the value worth coming back for.

Build distribution into the product

Too many founders keep building one more feature without answering how they’ll reach the target customer. A vague promise to do marketing can survive in a plan for months because nobody has made it specific enough to fail.

For a small software company, this announcement is a reason to spend a week getting specific. Pick one task customers already value and find out whether they try to do it in Claude. If they do, connect the smallest useful version and watch them use it. Find out where they get stuck and whether they want to return. That would give me much more to work with than another discussion about how big AI is going to be.

Then follow what happens after submission. The portal can show listing views and installs; your product needs to tell you whether those people complete useful work and become customers.¹ A hundred installs with no completed work wouldn’t persuade me to scale it. A few customers using it repeatedly would give me a reason to keep going.

You don’t need certainty that Claude will become the next app store. You need to know whether your customers are there and whether you can do something useful for them. If you wait until everyone agrees it’s a distribution channel, you’ll be competing with everyone who waited for the same answer.

Distribution is the most important feature. It deserves a place in what you’re building this week.