Build What’s Next on the ChatGPT Platform with Plugins | DevDay 2026
Build an MCP plugin around one useful user task, then use extensions to add a sidebar home, contextual side panel, or interactive file viewer.
- Add sidebar and side-panel entry points to an MCP widget; use model-context methods to share selections with ChatGPT.
- Expose a read-only MCP search tool for @ mentions, and package an onboarding skill plus settings read/update tools.
- Use Codex and App Plugin Creator to scaffold, test, and iterate; plugin.json-based submissions get automated checks.
The walkthroughs show plugin experiences in use and explain implementation points for extensions and events. · Best for: Developers building MCP plugins or integrating products with ChatGPT.
OpenAI presents ChatGPT plugin discovery, native extensions, MCP capabilities, and a simplified build-and-submit workflow. The talk includes demos and implementation details for developers.
Discovery and plugin growth0:25
- Plugins can be suggested in conversations as well as found in the directory.
- Discovery signals include post-install use, daily and weekly active usage, and retention.
- The speaker reports 4,000 plugins in the directory and increasing plugin use.
Extensions and interactive experiences5:55
- Plugin extensions bring interfaces into ChatGPT and Codex, including a sidebar home, side panel, composer actions, and interactive file viewers.
- Interfaces can pass selection and other context to the conversation and model.
- Demos include meeting notes and summaries, a 3D molecular-structure viewer, and a voice-driven Magic Path landing-page design.
- Examples from Atlassian, Canva, and Shopify show search, content creation, analytics, and commerce workflows.
Sign-in and building with Sites16:19
- Sign in with ChatGPT can connect a web app or plugin with ChatGPT identity; the speaker cites Supabase as an example.
- Eligible users can bring ChatGPT token entitlements to partner tools, funded by their ChatGPT subscription.
- Sites can host MCP servers; the speaker recommends starting with one user task and iterating with Plugin Creator.
Extension implementation20:20
- Extensions build on MCP and MCP Apps; add entry-point configuration to open widgets in the sidebar or side panel.
- Register supported file types to open files in an interactive widget.
- Expose a read-only MCP tool with the required search schema for composer @ mentions.
- Package an onboarding skill and provide settings through read and update tools.
Events and automations23:31
- The announced MCP event support lets a server notify ChatGPT of events that can trigger automations.
- Implement event/list to declare events and event/subscribe to receive a callback URL and signing secret; send a signed webhook when an event occurs.
Submission and getting started24:32
- The submission form is replaced by information in plugin.json, which Codex can generate from a codebase.
- Automated checks run on upload and provide feedback; deployed MCP server updates no longer require a new plugin submission.
- Start in Codex with an existing repository and ask App Plugin Creator to build a first-pass plugin or add an extension home.
Step by step
- Choose one user task your plugin can improve.
- Open the project in Codex and ask App Plugin Creator to review it and implement a first-pass plugin, or add a sidebar home to an existing plugin.
- Configure MCP entry points for the sidebar, side panel, and any supported file types; pass relevant interface context to the conversation.
- Add any needed composer search tool, onboarding skill, and settings read/update tools.
- Test the complete user experience, then iterate; use the updated directory submission checks when ready to publish.
Tools & resources mentioned
- App Plugin Creator — guides plugin creation and deployment.
- Codex — work with a repository to build and iterate on a plugin.
- Sites — host MCP servers and web apps.
- MCP / MCP Apps — tools and widgets used as the basis for plugin extensions.
- Sign in with ChatGPT — identity integration for web apps and plugins.
- plugin.json — contains plugin submission information and can reference onboarding configuration.
Notes are AI-generated from the video's transcript and may miss nuance.