30 秒でわかる
Plugin developers can place interactive MCP apps beside a conversation, expose content in composer search and register viewers for supported files, OpenAI’s Corey and Max Neuber say. They demonstrate meeting, research and design experiences, then explain onboarding, settings and event-triggered automation. They also describe a revised submission process using plugin metadata and automated checks.
- 1Discovery considers use after installation and retentionCorey describes how the directory and conversation suggestions can surface relevant plugins.
- 2MCP apps can appear beside the conversationExtensions add sidebar and side-panel entry points and share selection context with the model.
- 3Plugins can register file viewers and composer searchMax describes supported file types and a read-only search tool.
- 4Signed events can trigger an automationA server lists events, handles subscriptions and sends a signed webhook.
- 5Submission metadata and checks move into the upload flowMax describes plugin.json metadata, automated checks and MCP server updates.
ポイント
- Choose an entry point that matches where people need the tool: sidebar, side panel, composer or file viewer.
- Share relevant selection context between an interactive view and the conversation.
- Use onboarding and settings tools to help users configure a plugin.
- Test the full workflow and review the specification before submission.
チャプターごとに整えた全文を読む
How people discover plugins
Corey describes conversational suggestions, directory discovery and usage signals.
Corey from OpenAI’s developer experience team introduces changes to plugin discovery, interactive extensions and the process for building and submitting plugins.
ChatGPT can suggest relevant plugins in a conversation. Corey says discovery considers whether people use a plugin after installation, its active usage and whether users return.
He says the directory has 4,000 plugins and describes growth in installations, conversations using plugins and MCP tool calls as agents take on more involved work.
Interactive experiences in ChatGPT
Extensions expose interfaces, file viewers and composer search. Examples cover meetings, research and design.
Plugin extensions build on MCP apps and allow developers to put interactive experiences in a sidebar or side panel alongside a conversation. The interface and model can exchange information about what a user is viewing or selecting.
Plugins can expose search targets in the composer and register file types for interactive viewing. Corey says this lets a product participate in more of the ChatGPT workflow.
He demonstrates a meetings plugin in the sidebar that transcribes notes, gathers context from other systems and prepares a summary. Suggested action items can start new chats.
A life-sciences plugin example displays a molecular structure in a side panel. Corey says a selection in the viewer can be passed into the conversation so the researcher and agent share context.
In a Magic Path demonstration, a spoken request starts a landing-page design and opens it on an editable canvas. The example illustrates voice invoking a plugin while the user inspects the visual result.
Corey also describes examples involving Jira and Confluence content, a Canva presentation and Shopify store analytics. Each combines conversation with a product-specific interface or action.
Identity and building tools
The speakers describe sign-in, user entitlements and hosting a plugin server.
Sign in with ChatGPT can be added to a website or plugin. Corey says one partner saw more weekly installations after adopting it and describes eligible users bringing ChatGPT usage entitlements into a product.
The OpenAI developer plugin example has a sidebar home, starter prompts and a shareable DevDay schedule. Corey presents it as a reference for building an extension.
Sites can host MCP servers, he says, and Plugin Creator can help package and deploy a plugin. He recommends starting with one useful user experience and iterating.
Extension points and configuration
Max explains sidebar, file, search, onboarding and settings integration.
Max Neuber, who leads engineering for the plugin developer platform, explains that extensions use MCP and MCP apps. Existing widgets can be exposed through sidebar and side-panel entry points.
A file entry point declares supported extensions so a plugin can open matching files. For composer search, a read-only MCP tool accepts a query and returns matching objects to attach to a message.
Plugins can package an onboarding skill and expose settings using tools that return a schema and current values and save updates. ChatGPT renders the settings controls.
Events and a revised submission process
The platform supports event-triggered automation and automated submission checks.
Max describes support for an upcoming MCP event specification: a server lists available events, receives a subscription request and sends a signed webhook when an event occurs, triggering an automation.
He says plugin submissions have grown quickly and describes a revised process using plugin.json metadata and automated checks on upload. He also says MCP server changes no longer require a new plugin submission to go live.
The presenters encourage developers to try an existing project with Plugin Creator, then read the specification, test and iterate on the complete user experience.
同じチャンネルの記事
24:01EN
OpenAI· 2026年11月11日When agent work needs loops, subagents and a shared logRaj, Eric · OpenAIOpenAI’s Raj and Eric explain persistence, when to delegate and how they coordinate longer work.
22:37ENOpenAI· 2026年11月11日How GPT Live keeps speaking while an agent worksAlex, Erika · OpenAIOpenAI demonstrates simultaneous voice, background delegation and three ways to return updates.
21:10ENOpenAI· 2026年11月11日How OpenAI turned security findings into verified fixesIan, Kyle · OpenAIOpenAI engineers describe the scan, deduplication and patch verification loop behind Codex Security.
19:44ENOpenAI· 2026年11月11日What a managed agent harness has to do to keep workingCharlie, Steve · OpenAIOpenAI engineers explain remote sandboxes, long-running state and a human-approved flight rebooking demo.
27:09ENOpenAI· 2026年11月11日A six-week coding agent run and what it took to superviseOpenAI engineers describe a CI maintenance agent, long-running context and the cost of keeping it working.
20:48ENOpenAI· 2026年11月11日Sites adds personalized connectors and agent-friendly web appsKatia, John · OpenAIOpenAI demonstrates scheduled updates, per-user plugins and WebMCP in two Sites applications.