October 12, 2026
ChatGPT browser agents get context from annotations: an example with scene edits and expense updates
On Oct 12, the OpenAI Devs account showed in a 154-second video how annotations together with WebMCP give agents context for their actions. The video walks through an example of editing scenes and updating expenses in the ChatGPT browser.
OpenAI Developers
@openaidevs
Annotations together with WebMCP give your agents more context to act on. See how @kagigz uses custom annotation controls and ready-made prompts so users can edit scenes and update expenses in the ChatGPT browser.
· 74.3K views
Annotations and WebMCP give the agent the context it acts on. @kagigz showed this in ChatGPT in the browser: custom annotation controls and ready-made prompts help users edit scenes and update expenses.
What this changes for builders. Ready-made prompts in the interface give users a starting point instead of an empty field. If you are building an agent that works inside a page, prompts and element markup become part of the product.
So far this is one walked-through scenario, with no measurements or figures for speed or accuracy.