Hey, this is a known issue in versions 2.3.18+ and Early Access builds.
Key point: Browser automation in Cursor works through built-in tools and should NOT show up as a separate MCP server in Settings > MCP. When you see “No MCP resources available” after asking “@Browser is there browser tools?”, that’s expected because Browser isn’t MCP, it’s a built-in feature.
That said, in your version (2.4.0-pre.48 Early Access) there’s a regression and Browser automation may not work at all. The team is working on a fix, and the ticket is tracked here: Browser MCP Tools Not Available
Workaround: downgrade to 2.2.44, where Browser works reliably:
Hey, good question. Built-in Browser tools work like this:
When tools get called:
The AI automatically gets a list of available Browser tools (navigate, click, screenshot, console, network, etc.) when the Agent is running with browser capabilities.
The AI decides on its own whether to use them based on your request context.
For example, if you say “open google.com in the browser” or “test this page”, the AI sees tools like browser_navigate and browser_screenshot and decides to use them.
How it works technically:
Tool descriptions are passed to the AI via the system prompt, so it knows what tools are available and what each one does.
The AI analyzes your request and decides which tools it needs, same as with other tools like file_edit or terminal_execute.
You see it in the chat when the AI shows something like “Using browser_navigate…” before it acts.
Important: in your version (2.4.0-pre.48 Early Access) this is broken due to a regression, so even if the AI tries to call the tools, they won’t work. That’s why downgrading to 2.2.44 is used as a workaround.
So far, it’s possible to re-@Browser, but the pre-release version of cursor is really too unstable. It didn’t work last week, but with this week’s update, it just started working.
