You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Browserstack integration with Cline in Windows is not working #69
Can you try running this npx @browserstack/mcp-server@latest in powershell ?
Expected output is actually nothing, the server will start but won't print anything to the console.
I can able to run the application with command line interface with below command.
npx -y @browserstack/mcp-server@1.1.5 --BROWSERSTACK_ACCESS_KEY= --BROWSERSTACK_USERNAME=
Once the application is starts I need to configure this MCP server with Cline and Cline tool should be able to integrate with running MCP server. When Cline is try to connect this MCP I am getting above mentioned issue.
Looks like from Cline when try to connect the running MCP (bowserStackServer) its trying to make some changes in my VS Code folder file that's where I am getting Operation Not permitted issue.
I’m seeing what appears to be the same issue with @browserstack/mcp-server 1.3.3 on Windows, launched by Codex Desktop using Node.js and npx.
In my case, the working directory is writable, so the server starts successfully but creates a literal NUL file in the repository root. Git then shows it as an untracked file. Deleting it only helps until the server starts again.
The file contains these startup messages:
Launching BrowserStack MCP server, version 1.3.3
Creating BrowserStack MCP Server, version 1.3.3
The installed dist/logger.js appears to explain the behavior. Its default, non-development branch uses:
The intent is to discard logs, but this logging path produces a real file named NUL on my system. That would also explain the original report’s EPERM when the working directory is protected.
Could the default branch disable logging directly—for example, with pino({ enabled: false }) and no file transport—instead of opening a null-device path? Any enabled diagnostic logging should also stay off stdout because stdout carries the MCP protocol.
A useful Windows regression check would start the server with NODE_ENV unset in an empty writable directory, then verify that it initializes without creating NUL or another log file. Testing a non-writable working directory would cover the original failure too.
Activity
pulkitsharma07 commentedon Jun 16, 2025
Can you try running this
npx @browserstack/mcp-server@latestin powershell ?Expected output is actually nothing, the server will start but won't print anything to the console.
uthircloudnative commentedon Jun 16, 2025
I can able to run the application with command line interface with below command.
npx -y @browserstack/mcp-server@1.1.5 --BROWSERSTACK_ACCESS_KEY= --BROWSERSTACK_USERNAME=
Once the application is starts I need to configure this MCP server with Cline and Cline tool should be able to integrate with running MCP server. When Cline is try to connect this MCP I am getting above mentioned issue.
Looks like from Cline when try to connect the running MCP (bowserStackServer) its trying to make some changes in my VS Code folder file that's where I am getting Operation Not permitted issue.
korggy commentedon Sep 13, 2026
I’m seeing what appears to be the same issue with
@browserstack/mcp-server1.3.3 on Windows, launched by Codex Desktop using Node.js andnpx.In my case, the working directory is writable, so the server starts successfully but creates a literal
NULfile in the repository root. Git then shows it as an untracked file. Deleting it only helps until the server starts again.The file contains these startup messages:
The installed
dist/logger.jsappears to explain the behavior. Its default, non-development branch uses:The intent is to discard logs, but this logging path produces a real file named
NULon my system. That would also explain the original report’sEPERMwhen the working directory is protected.Could the default branch disable logging directly—for example, with
pino({ enabled: false })and no file transport—instead of opening a null-device path? Any enabled diagnostic logging should also stay off stdout because stdout carries the MCP protocol.A useful Windows regression check would start the server with
NODE_ENVunset in an empty writable directory, then verify that it initializes without creatingNULor another log file. Testing a non-writable working directory would cover the original failure too.