Feature request: Support remote browser connection via MCP or daemon mode

[AzzzGoodFish]

Use case

I run Claude Code on Linux, but my Playwright MCP Server runs on a Windows machine. This setup allows me to see browser actions in real-time on my Windows desktop while Claude Code controls it remotely.

Currently, dev-browser is a local CLI tool — it installs and launches Chromium on the same machine. This means if I install it on Linux (where Claude Code runs), the browser opens on Linux and I can't observe it from Windows.

Proposal

It would be great if dev-browser supported a remote mode, for example:

1. Daemon/server mode — Run dev-browser serve on the Windows machine, exposing an API that remote clients can call. Claude Code on Linux would send scripts to the Windows daemon for execution.
2. MCP server mode — Similar to @playwright/mcp, expose dev-browser's capabilities as an MCP server, so Claude Code can connect to it as a tool provider.

This would combine dev-browser's strengths (sandboxed scripts, batched multi-step operations, fewer round-trips) with the ability to run the browser on a different machine for real-time observation.

Current workaround

Using @playwright/mcp directly, which supports remote connections but requires more round-trips per task.

[SawyerHood]

this is a good feature request!

I am 100% onboard with 1) (having a way to serve).

2 i need to think about a bit more, but it would be useful for using dev-browser outside of traditional coding agents, i'm gonna have to noodle on that a bit more

[AzzzGoodFish]

Thanks for the quick response!

Agreed, mcp tools might not be necessary.

The skill + CLI pattern is working well in practice, and a ‘serve’ mode would fit naturally into that workflow.

Looking forward to it!

[coreyward]

I feel like this should be possible via SSH tunnel w/o having to do anything extra on the remote, but I think dev-browser is binding to localhost preventing it from working properly.