Build1 publisher2 min readPublished
OpenAI opens the plugin system behind ChatGPT's own features to outside developers
OpenAI opened the plugin system behind ChatGPT's own features to outside developers at DevDay and launched a 32-partner enterprise marketplace. For teams that build internal tools, ChatGPT is now both a place their integrations can run and a product that competes with them.
The Engineer · Build desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Through the MCP Events specification and a feature called Team Tasks, ChatGPT can start automated workflows when something happens in a connected app.
- ChatGPT now answers @ChatGPT mentions in Slack and Microsoft Teams channels, threads and direct messages, and the person mentioning it needs no ChatGPT license.
- Plugin Extensions let developers put interactive panels beside a conversation and build viewers for the file types their plugins support.
- A Sign in with ChatGPT option lets users draw on their ChatGPT quota inside 16 third-party tools.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision Internal-tool teams now have to pick which layer they own: a connector that any ChatGPT surface can call, or an interface their users open directly and OpenAI does not host.
- exposure Slack and Teams members without a license can now reach tools an admin connected to ChatGPT, subject to permissions, so the scope an admin grants at connection time applies to everyone in the channel.
- constraint MCP Events is still a proposed specification, so event-driven workflows built on it today depend on a contract that can still change.
- precedent OpenAI shut down shared chats in July to keep ChatGPT simple, so a team process built around Space or Pages could be withdrawn the same way.
A plugin built on this system answers to more than one principal. Each user installs it and approves its access, on every ChatGPT plan [4]. If a plugin reaches an internal system through one shared service account, every teammate sees the same rows. That design breaks inside a user-created Site, where team members use the same plugin with their own data and permissions [5].
Distribution runs through OpenAI's directory. OpenAI says a new Plugin Creator makes building easier, and that a revamped submission process brings clearer feedback and better ranking in the directory [3]. Shareable profiles let users show their Sites and plugins so others can find and reuse them [17]. According to The Decoder, OpenAI recently hired Patreon co-founder Sam Yam to lead a new creator product division [6].
Event triggers change who starts the work. A prompt has a user behind it. Under MCP Events and Team Tasks, the starting signal is something that happened in a connected app [7]. Before wiring a production workflow to that, I'd want the delivery semantics written down. I'd want to know what happens when an app fires the same event twice, and what happens when it fires while the connector is down.
The competing products are already in the release. Pages is a document type built for humans and AI agents to write in together [15]. Collaborative Slides arrive in the coming weeks and export to PowerPoint or Google Slides [12]. So OpenAI has put an export to the incumbents' formats inside the product built to compete with them [12][13]. The Decoder writes that the announcements put OpenAI on a collision course with at least Slack, Notion and Google Workspace, and ultimately even Microsoft Office or Windows [13].
My view assumes a team whose internal tools are mostly an API with a form in front of it. For a tool like that, I think shipping the function as a plugin is the right choice. A tool connected to ChatGPT can also be reached from Slack and Teams through @ChatGPT mentions [8][9]. If a tool's value is its interface or the workflow state it keeps, I'd keep building it outside ChatGPT. OpenAI now ships its own shared workspace, Space, where the Dot assistant pulls context from shared material and hands tasks off to other agents [14].
What to watch
- Whether the MCP Events proposal is adopted in the form ChatGPT implements, or changes after ChatGPT ships it.
- The admin controls OpenAI publishes for @ChatGPT in Slack and Teams, especially how per-user permissions apply to members without a license.
- Whether directory ranking and the 32-partner marketplace favour OpenAI's own Space, Pages and Slides features over third-party plugins.