According to ExtremeTech, OpenAI has officially opened submissions for its new ChatGPT App directory, letting developers publish tools directly to the platform using a new Apps SDK. Approved apps will be listed in a directory accessible through the tools menu or a direct link, and users can launch them by typing “@” in a chat. The system enables software to perform real-world tasks like ordering groceries or creating slides from an outline. OpenAI is also testing an automatic recommendation engine that suggests relevant apps based on conversation context. This launch coincides with other ecosystem plays, like Anthropic’s Model Context Protocol (MCP), and OpenAI now requires app submissions to include details on MCP connectivity.
The Conversational OS Is Here
Here’s the thing: this isn’t just a new feature. It’s a fundamental shift. For years, we’ve talked about AI assistants being “conversational interfaces.” But now, OpenAI is building the actual operating system for that concept. Typing “@” to summon a specialized tool mid-chat is a brilliantly simple mechanic. It keeps you in the flow. You’re not closing a tab, opening a new app, and pasting data. The action lives inside the conversation.
And that recommendation engine they’re testing? That’s the secret sauce. If ChatGPT can reliably suggest the Adobe Photoshop tool when you mention editing a photo, it stops being a chatbot and starts being a proactive concierge. The value skyrockets. Suddenly, the AI isn’t just answering questions—it’s orchestrating a suite of services to solve your problem.
What This Means For Everyone Else
For developers, this is a huge new channel. But it’s also a walled garden. Getting approved for OpenAI’s directory is now a goal, with MCP compatibility part of the checklist. That pushes the industry toward a specific standard. For users, it means ChatGPT is about to get a lot more powerful and, frankly, a lot more cluttered. How do you discover a useful app in a sea of them? The recommendation system will be critical.
For the market, this is OpenAI’s clear shot across the bow of every other AI company and even traditional software platforms. They’re not just building a better chatbot; they’re building the platform where work gets done. Anthropic’s MCP is a smart, open counter-play, trying to ensure interoperability so developers aren’t locked into one ecosystem. This is the beginning of the platform wars for AI.
Basically, we’re watching the early, messy formation of a new software layer. It’s exciting, but it also raises big questions. Who curates these tools? How are they secured? And what happens to all the standalone apps these mini-tools might replace? One thing’s for sure: the AI landscape just got a lot more interesting.
