Users want the tools they need right at their fingertips. They don’t want to leave their way of working to find and use new capabilities. AI companies quickly realized that it must be easy for users to integrate third-party services with agents. Those third-party services are, in fact, the doors to the external world that agents use to fulfill users’ needs. And MCP, the Model Context Protocol, is right at the center of the connection between AI agents and third parties. How, then, is MCP becoming the standard way to find and use third-party tools? Stay with me to learn more.
This article is brought to you with the help of our supporter: Speakeasy.
Turn your API platform into an AI platform with Gram by Speakeasy. Create tools from OpenAPI, curate into custom toolsets, and deploy hosted MCP servers.
“No AI without APIs” doesn’t make sense. At least in a world where agents don’t need to communicate with the outside. You can have all the APIs you want. If your AI system can’t reach them, they’re useless. Until a certain point, AI evolved inside its own internal world. There was always this tension building to breach the AI’s walls and let third-party data flow through. However, all the efforts couldn’t gain mass adoption. Different companies were releasing their own solutions, always with limited success. Until one of those companies decided to release its approach as an open-source project. And, this is pretty much how the Model Context Protocol came to life. Its adoption went through the roof almost instantly, and the number of MCP servers grew without control. Suddenly, users went from not being able to connect AI to the outside to not being able to find the right tools to use. With so many MCP servers available, how would you know which ones were right for you? Someone had to do something. The people behind MCP could have kept quiet, but instead, they decided to offer a solution to the problem.
The solution MCP came up with is as open as the protocol itself. The MCP registry is, according to the official website, “like an app store for MCP servers.” Its goal is to serve as the “authoritative repository for publicly available MCP servers.” It was launched in September 2025 and, two months later, it already had over 8,000 MCP servers listed. This kind of growth is only possible because everyone is publishing their servers there. Everyone wants their servers to be listed on the authoritative MCP registry. I mean, wouldn’t you? Since there is no manual curation process involved, the only thing stopping growth is the lack of new MCP servers being created, which isn’t happening anytime soon.
On top of that, there is little to no friction in publishing servers. The MCP registry is not only a Web interface but also an API. What that means is that any tool that lets users create and deploy MCP servers can immediately publish them on the registry. Zero effort. Zero friction. This approach is brilliant. They could have simply centralized the interface and made MCP owners go to a Web interface to publish their servers. However, by making the publishing available through an API, they’re extending the capability to any tool that wishes to offer it. Not only are MCP owners able to publish their servers right from their tool of choice, but end users are also able to find their favorite MCP tools right from inside their agent interface. Suddenly, you have a platform ecosystem in place. On one side, you have MCP producers, and on the other, you have all the end users consuming the tools. Their opportunity is to become the middle layer between MCP producers and consumers. They can even launch a marketplace if MCP popularity continues to grow. That is, if they’re interested in exploring that path.
The path to becoming the de facto MCP marketplace is full of challenges. One that is probably the most difficult to address has to do with quality. Having a lot of published MCP servers is great, right? Not until searching for tools becomes painful. To overcome the potential growing lack of quality, MCP will have to curate MCP servers. Or, at least, mark some of them as “verified,” or something similar. For end users, retrieving too many results that look like junk or spam becomes a pain and breaks the trust they have in the registry. Not trusting the authoritative MCP registry makes end users go elsewhere. And, if one side of the platform ecosystem collapses, then the other side (the MCP server owners) will also go elsewhere. There’s a fine balance to be maintained, and its main driver, quality, is definitely something to consider going forward.
The registry opportunity is here for MCP to grab it. If they don’t, someone else will, for sure, if MCP popularity continues to grow. Yes, there are challenges. But the benefits of becoming the most important MCP-related service can’t be ignored. I hope the registry grows to become a healthy platform ecosystem where different kinds of businesses can then thrive.
