[Question] browserType.launch timeout and multiple connections

[qe360-vrucenzon]

Hello!

I'm testing out the Playwright MCP (0.0.27) containerized on ECS. I have another back-end server on ECS that connects to the MCP over SSE, and I also have an MCP inspector running locally. The back-end server can talk to the MCP no problem, but I noticed that when I use the MCP inspector (ex. running browser_navigate), subsequent requests made by the back-end server to the MCP result to browserType.launch timeouts (which happens when browser_navigate is called):

When I disconnect the MCP inspector, the back-end server can talk to the the MCP without problems again.

Is this expected behavior for this scenario? Or am I just missing some setup to support the back-end and MCP inspector connecting to the MCP over SSE at the same time?

If there's any other information I need to provide, please let me know. Thank you.

[pavelfeldman]

Do you want your connections to use the same browser or different browsers?

[qe360-vrucenzon]

Different browsers.

I currently use these args: --isolated, --no-sandbox

[pavelfeldman]

It will be launching separate browsers, which will result in elevated resource consumption, but it should work. I'll land a change that reuses the same browser with isolated contexts in sse mode, but technically existing setup should work. I'll add a test as well.

[qe360-vrucenzon]

Right, that makes sense. Ok I'll test it out once available, thank you very much.

[cloudycotton]

I think I'm facing the same issue in #482

[pavelfeldman]

Next version will provide a more readable error and suggest using --isolated.