Until this week, a ChatGPT plugin was something the model called. From now on it can be something the user opens. OpenAI has published plugin extensions, which let outside developers hook their software into the interface surfaces OpenAI uses to build ChatGPT itself: the sidebar, the composer and the file viewers.

The list is specific. A sidebar app gets its own entry in ChatGPT’s global sidebar and opens fullscreen. A conversation panel opens beside a chat so a user can work in the app and talk to the model at the same time. A file viewer registers the extensions it supports and opens matching files in its own interface, with reading, live updates and saving handled together. Plugins can declare their own settings page inside ChatGPT, deep-link to a specific item, and be mentioned from the desktop composer to pull in content.

The mechanism is MCP, extended

The capability is delivered as OpenAI MCP Extensions, described in the documentation as a formal, language-agnostic API extending the Model Context Protocol and the MCP Apps specification. There are TypeScript and Python SDKs, a published protocol specification, and a demonstration plugin called Bits & Bolts that exercises each extension.

A separate piece of the same release adds support for the proposed MCP Events specification, which lets a plugin start an automation when something happens in a connected app — a new item on a project board triggering ChatGPT to read the linked documents and draft a plan, in OpenAI’s own example. That turns plugins from things a user invokes into things that run on their own.

A modular shelving unit with labelled compartments
Extensions cover sidebar apps, panels, settings and file viewers. Illustrative photo. Andreas Näslund · pexels · Pexels License

The rollout is uneven. Plugin extensions on the web are described as coming soon to ChatGPT Free and Go users, and composer mentions are only in the desktop app. Existing plugin functionality is unchanged. On the distribution side, OpenAI added a Plugin Creator, a redesigned submission flow with clearer feedback, and improved ranking in the directory and in conversations. Business, Enterprise, Healthcare and Edu customers can host supported plugins inside the Sites they build.

The number underneath it

A developer platform is worth what its distribution is worth, and OpenAI put a figure on that. In the DevDay recap the company describes opening ChatGPT “as a shared surface where humans and agents can collaborate and where developers can directly launch new native experiences to our collective 1.2B weekly users”. The figure is OpenAI’s own, disclosed by OpenAI, and has not been independently verified.

It is also the argument. A sidebar slot in front of 1.2 billion weekly users is a distribution channel of a kind that has historically been owned by operating systems and app stores, and the terms here are set by one company that also supplies the model, the billing and the ranking.

A crowd of people crossing a busy street
OpenAI puts ChatGPT's weekly reach at 1.2 billion people. Illustrative photo. Jimmy Liao · pexels · Pexels License

What developers are actually agreeing to

OpenAI’s documentation is clear that users choose which plugins to use and approve the access each one receives, which puts the consent boundary in the right place. What it does not address is the platform boundary: the same directory ranking that surfaces a plugin can bury it, and the same surfaces that host third-party apps host OpenAI’s own — Pages, Space, Codex, the Meetings plugin all sit in the sidebar the extensions API now shares.

What to watch

The first real test is the ranking. OpenAI says it has improved recommendations in the directory and in conversations without saying how either works, and an app store’s economics are decided by that answer. The second is whether MCP Events is adopted as a standard or remains an OpenAI dialect: it is a proposed specification, and a plugin ecosystem that runs on one vendor’s extension of an open protocol is not the open ecosystem the protocol was for.