On September 29, 2026, OpenAI announced at its DevDay developer conference in San Francisco that ChatGPT Plugin Extensions had officially launched. For the first time, developers can build full application-level experiences inside the ChatGPT interface with independent sidebars, conversation panels, and file operation entry points. The first three companies to launch—Canva, Figma, and Adobe—integrate through different technical paths.
Three Access Modes, Three Technical Paths
Plugin Extensions is a development framework composed of multiple interface layers. Officially called “OpenAI MCP Extensions,” it is built on the Model Context Protocol (MCP)—an open standard designed to connect AI models with external tools.
Developers can build several types of interfaces: sidebar apps that open full-screen, conversation side panels that remain visible while a chat is in progress, file viewers and editors that support a product’s proprietary formats, and structured forms triggered by @-mentions in the message input box. This “bidirectional context sharing” mechanism allows apps and AI models to exchange state, rather than merely making one-way calls.
The three initial partners chose very different entry points:
- Canva chose the sidebar tab, letting users preview and create designs directly in the sidebar while conversing with ChatGPT.
- Figma uses the desktop “Composer mention” feature, allowing users to search for and reference Figma files directly while composing a message.
- Adobe uses a File Handler to embed editing capabilities for Acrobat and Photoshop files directly into ChatGPT, so users can work with documents without leaving the conversation interface.
These three modes cover three different depths of workflow integration: lightweight preview, contextual reference, and deep editing. OpenAI used leading creative tools to open up all three paths, demonstrating the framework’s boundaries to future developers.
The Internal Logic of Platformization
ChatGPT has strong user stickiness, but most professional workflows still happen in other applications. Users draw in Figma, make posters in Canva, and sign contracts in Acrobat; previously, ChatGPT could only offer text suggestions at the edges of these workflows.
Plugin Extensions changes where that boundary sits—not by pulling users back from third-party tools into ChatGPT, but by letting the core interactions of third-party tools happen directly inside ChatGPT. This logic is highly similar to the platformization paths of Apple’s App Store and WeChat Mini Programs: the party holding the traffic gateway lowers the cost of developer integration, coalesces the ecosystem, and gives users the experiential value of “not having to switch apps.”
OpenAI published the technical specification in a GitHub repository and provided two SDKs, for TypeScript and Python, while also updating the submission process and recommendation algorithm for the plugin directory. These infrastructure moves show that this launch is aimed not only at Canva, Figma, and Adobe—it is an invitation to the broader developer community.
Real-World Constraints: Not a Full-Scale Rollout
The Composer mention feature is currently limited to the desktop app, with no support on the web yet. The web sidebar extension for ChatGPT Free and Go plan users is marked as “coming soon,” with no specific timeline. Workspace administrators for enterprise accounts can restrict users from installing plugins, which means that in enterprise settings, the pace of Plugin Extensions adoption will be constrained by IT control policies.
These limitations show that the framework is in place, but availability remains tiered. Users who can experience the full functionality are currently concentrated among paying users on the desktop client. For the larger share of free web users, this launch brings limited practical change to their experience in the short term.
Head-to-Head with Microsoft Copilot
The timing of the Plugin Extensions launch cannot be understood apart from the competitive context with Microsoft. Microsoft Copilot has been deeply integrated into the Office 365 ecosystem for years; Word, Excel, PowerPoint, and Teams all have built-in Copilot entry points, and enterprise users already have a relatively mature AI-assisted path for workflows within the Microsoft ecosystem.
Canva, Figma, and Adobe happen to be among the most important creative tools outside Microsoft’s ecosystem. Canva focuses on small and medium-sized businesses and content creators, Figma is the core workbench for professional designers, and Adobe covers professional image processing and document management. By choosing these three as its first partners, OpenAI is establishing a presence first in design workflows that Microsoft Office cannot reach.
From this perspective, ChatGPT Plugin Extensions is a battle to seize the “AI gateway to the enterprise creative toolchain,” not merely a feature iteration.
The MCP Protocol: OpenAI Is Playing a Bigger Game
OpenAI chose to build Plugin Extensions on top of MCP (Model Context Protocol), an open protocol rather than a proprietary OpenAI standard. MCP is designed to standardize connections between AI models and external tools, and multiple AI vendors, including Anthropic, have participated in advancing it. By choosing to build the Plugin Extensions system on this open standard rather than starting from scratch with a proprietary protocol, OpenAI lowers the barrier to developer integration—tools that already support MCP can, in theory, integrate with the ChatGPT extension framework with only a small amount of additional work. This choice helps rapidly expand the developer ecosystem in the short term, but in the long run it also means OpenAI has given up some opportunities to build barriers through proprietary protocols.
Verdict
The technical feasibility of ChatGPT Plugin Extensions has been validated in practice by Canva, Figma, and Adobe—the most valuable signal from this launch: the framework is not just documentation; it can already run in mainstream production tools.
But whether the platform can succeed depends on two things: whether developers are willing to come, and whether users stay. The currently visible limitations—desktop first, web pending, enterprise controls layered on top—show that OpenAI is still using a “precision rollout” approach for a cold start rather than a full-scale rollout. The real moment of validation will be six months from now: how many plugins from independent teams are in the developer directory, and what proportion of enterprise users’ daily active workflows involve Plugin Extensions.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接