Summary:
The browser_snapshot tool sometimes returns a response where the model only receives the metadata part (e.g. Unsupported content type “metadata”: {“viewId”:“…”,“title”:“…”,“url”:“…”}). The accessibility tree (YAML with role, name, ref for elements e21, e22, … e78, etc.) is visible in the Cursor UI to me but is not included in the tool result text that the model receives. So the model cannot extract page content (e.g. job descriptions from ATS pages) even when the snapshot is full in the UI.
Request:
Is it possible to pass the full browser_snapshot response—including the accessibility tree (all elements with name fields)—to the model as text, so that workflows that rely on extracting content from the snapshot (e.g. job description extraction from JS-rendered job boards) can work without requiring me to paste content manually?
Context:
Used in a job-search workflow: fetch job URL → minimal HTML → browser_snapshot to get JD from Ashby/Lever/Greenhouse etc. The snapshot in the UI contains the full JD in name fields; the model only sees metadata and writes a stub asking the user to paste.