I’ve been using Playwriting MCP via a container (not from Cursor’s MCP directory) and it also works completely unreliably in 1.0. It rarely can do longer sequences of commands while in 0.5 it worked much more reliably.
Not sure why the MCP “reload” icons were removed.
There is no logging, nothing to see what the orange dot means. When I stop the MCP it goes red but then it’s stuck at orange until I reload the Cursor window, just to turn orange after a (very) few interactions with the MCP. Other MCPs work OK.
Even when the MCP status dot is red, I can see it making the first request in the server logs but then it’s stuck.
I was having the same issue with the following printed in the logs:
2025-06-17 13:41:13.890 [info] ight: Starting new stdio process with command: npx @playwright/mcp@latest
2025-06-17 13:41:13.890 [error] ight: Client error for command A system error occurred (spawn npx ENOENT)
My workspace has Node installed via Mise which caused the issue for me. I’m able to work around this by creating a separate Bash script which loads the Mise environment in it and then runs the npx command.
Playwright is a godsend when it works. But it gets fussy several times a day.
This is probably overly heavy handed but I’ve had to keep resetting & restarting mine with this script. Not sure if something with how I use it is strange, but I am constantly getting profile related errors.
It be incredible if playwright MCP had its own tool to manage itself more.
It is wonderful and closes the loop between ai coding and browser testing beautifully. But as an automation newbie, it does get confusing, especially with all of the ways it can launch a browser.