The Agent just cannot use the Browser MCP. I’m not even sure where the MCP went in the MCP settings as I don’t see it anymore. There is a icon in agent chat and when I hover over it it says “Connected to Browser Tab”. But when the agent is prompted to open a page it just fails.
Hey, thanks for the report. This is a recurring issue with the built-in browser tools. Try this:
Update Cursor. You’re on 2.4.22 (Jan 26), which is pretty old now. There have been several fixes for Browser MCP since then. The team fixed a similar bug around 2.3.18+. Go to Help > Check for Updates.
Toggle Browser Automation. Go to Settings > Tools & MCP > Browser Automation, turn it off, wait about 10 seconds, then turn it back on and set it to “Browser Tab”. After that, fully restart Cursor, not just Reload Window.
After restarting, open a new Agent chat and try again.
Note: cursor-ide-browser is the built-in MCP server. It won’t show up as a separate entry in the MCP config list. It’s fully controlled by the Browser Automation toggle in Settings > Tools & MCP.
If it still doesn’t work after updating and toggling, please send:
The Request ID from the failed chat (three dots in the top right > Copy Request ID)
Any errors from Help > Toggle Developer Tools > Console when the agent tries to use the browser
Using nightly build I also have issues with browser mcp. when asking “auto” mode to explain whats going on this is the answer I got:
Blocker: In this environment, when I call browser_snapshot, the tool result only shows metadata (viewId, url, locked, etc.). The snapshot body (the tree and refs) is not present in the result I get, so I never see the refs needed for browser_click. So the workflow is correct, but I can’t complete it here until the snapshot response includes the actual snapshot content (or refs are provided some other way). The rule file records this workflow and notes that if snapshot content/refs are missing from the tool result, ref-based actions can’t be done until that’s fixed.
icon in agent chat and when I hover over it it says “Connected to Browser Tab”. But when the agent is prompted to open a page it just fails.