ChatGPT Opens an App Directory, Turning the Chatbot Into a Platform
OpenAI now lets developers submit apps built on MCP for review and listing inside ChatGPT, a first step toward an app store with 800 million weekly users.
Key takeaways
- Developers can submit Apps SDK apps for review; approved ones appear in a searchable directory at chatgpt.com/apps.
- Apps are built on the Model Context Protocol and render custom interfaces inside the chat window.
- At launch developers could not yet sell digital products, so only physical goods could be sold through apps.
In December 2025 OpenAI opened submissions for third-party apps and launched a browsable directory inside ChatGPT. It is the clearest sign yet that the company wants its chatbot to be an operating layer, not just an answer box, and it arrives with a very large audience: 800 million weekly users, the figure Sam Altman gave at DevDay in October.
What happened
OpenAI unveiled the Apps SDK at DevDay, letting partners such as Zillow and Spotify run inside conversations. In December it took the next step: developers can now submit finished apps through OpenAI's developer platform, where they go through a quality-and-safety review. Approved apps are listed in a directory that users can browse and search, at chatgpt.com/apps.
Apps are summoned in two ways. A user can name one with an @ mention, or ChatGPT can decide that an app is relevant to what the person is asking and offer it. Popular apps may get featured placement or be surfaced automatically, which makes ranking inside the model's own routing logic the new distribution battleground.
How an app gets into a conversation
Under the hood, apps use the Model Context Protocol (MCP), the open standard for connecting models to tools and data. The developer exposes capabilities through an MCP server and attaches a custom interface, such as a map, a listing carousel or a slide editor, that ChatGPT renders in the chat. OpenAI also open-sourced an example set and a UI library for chat-native components.
Why it matters
Apple and Google own the app store model on phones; OpenAI is trying to build a second one where the interface is a conversation. If users get used to saying "book it" or "find me an apartment" without leaving ChatGPT, the app a developer ships becomes a tool call rather than an icon. That shifts power: discovery depends on the model's routing, and the review process becomes a gate comparable to App Review.
The commercial rules are still thin. At launch, developers could not sell digital goods through apps, so only physical products were supported, and gambling and adult content were banned. Apps also have to offer clear value beyond what ChatGPT already does natively. No developer revenue-share terms were described in the coverage reviewed.
What developers are being asked to build
OpenAI's own examples show the intent: grocery ordering, slide-deck building, apartment booking, shopping, data visualisation, and restaurant, travel or appointment booking. These are the tasks where a chat can gather the intent and an app can complete the transaction. The pitch to developers is that a single MCP server and a small interface layer put their product in front of a very large conversational audience without a separate install.
The submission rules are explicit. Apps must meet guidelines on safety, privacy and quality, and they must offer value beyond what the model does on its own, which rules out thin wrappers that merely re-skin a prompt. Gambling and adult content are banned. Because there is no developer payment model described yet, early apps are mostly lead generators, booking funnels and utilities for businesses that already make money elsewhere.
The strategic comparison is to the early iPhone App Store, but with a twist. On a phone, the user picks an icon. In ChatGPT, the model picks, which means that how well an app description and tool schema match the way people phrase requests will matter more than a splash screen or a store listing. Several marketing and app-store-optimisation firms have already started publishing guides on how to rank inside the directory, a sign that the discovery game is forming before the directory has mature analytics.
What to watch
- When OpenAI enables digital-goods sales and what cut it takes.
- Whether featured placement turns into a paid or ranked ad surface.
- How competitors respond: MCP is an open standard, so the protocol may matter more than any one directory.
- Quality control as the first wave of apps meets real user traffic.
