OpenAI has introduced plugin extensions that let developers build applications directly into ChatGPT and Codex, adding new ways for plugins to appear across the ChatGPT interface.
Developers can place a plugin in the sidebar, add an interactive panel beside a conversation, or create a viewer for specific file types. The system is based on OpenAI MCP Extensions, which connect plugins to parts of the ChatGPT experience including the sidebar, composer, and file viewers.
Under the new setup, developers can register support for a range of extension types with a small amount of SDK code. These include sidebar apps, conversation panels, plugin settings, file viewers and editors, display modes, deep links, model-app context sharing, composer mentions, and rich forms for structured user input.
Plugin extensions on the web will be available soon to ChatGPT Free and Go users, while composer mentions are limited to the ChatGPT desktop app. Existing plugin functions will remain unchanged.
Developer tools
OpenAI has published developer tools in TypeScript and Python, along with a protocol specification that defines the language-agnostic application programming interface. Developers can use these resources to install extensions, connect and test a plugin, and prepare it for distribution.
One option lets developers add an entry to the global sidebar so users can open an app in full screen from within ChatGPT. Another allows a plugin to appear in a conversation side panel, keeping the app visible alongside an active chat.
File handling is another part of the rollout. Developers can set a plugin to open when a user selects a supported file extension, with reading, live updates, and saving handled in the same interface.
OpenAI also outlined richer forms that go beyond standard MCP forms. These can collect structured input or image-based selections from users and return the response to a tool.
Examples in use
OpenAI pointed to early implementations from Canva, Figma, and Adobe. In those examples, Canva uses sidebar tabs for design work and previews, Figma uses composer mentions so users can search for and insert files in the desktop app, and Adobe uses file handlers for editing through Acrobat and Photoshop within ChatGPT.
OpenAI is also offering a demonstration plugin called Bits & Bolts, describing it as a broad test environment for trying extension features as they are added. The example is intended to show developers how the new interface components work across different parts of ChatGPT.
The move reflects a broader effort to make ChatGPT a host environment for third-party software rather than just a conversational interface. By giving plugins defined places within the product, OpenAI is providing developers with a more formal way to build services that fit into the chat workflow.
Several of the additions aim to reduce friction in navigation and context. Deep links can send users directly to a specific page or item within a sidebar app, while model-app context sharing is designed to keep ChatGPT and an external MCP app aligned through bidirectional context exchange.
OpenAI said support for these extensions can be declared with only a few lines of SDK code. It added that OpenAI-registered MCP servers require multi-round-trip requests, while the SDK documentation includes a legacy input helper for direct MCP connections.
The rollout gives developers a new way to distribute software through ChatGPT while preserving existing plugin behaviour.