Localhost is trapped on your machine
You can read everything Claude Code does from your phone. You can't see any of it.
You can read everything Claude Code does from your phone. You can’t see any of it.
A terminal is text, and text travels anywhere a connection goes. You SSH in over Termius or Blink and the whole session streams to your phone, every line, every diff. The reading half is solved.
Then it rebuilds your homepage, and you have no way to look at the page. The result of frontend work is a rendered page, sitting at localhost:5173 on your machine. localhost means that machine. Your phone is a different machine, so the one output you most need to see is the one that doesn’t come with you.
To actually look at it, you open a tunnel and put your half-built site on a public URL you have to manage. Or you deploy the branch just to look, and wait for the build. Or you wait until you’re back at the desk. A few terminals have a preview built in, and the ones I tried wanted extra setup on my Mac I didn’t want.
And you have to look, because the agent is confidently wrong about UI work. It tells you it finished the task. It changed something. Sometimes it even checks its own work with Playwright and still reports success. Then you look, and it built something different from what you asked, or it’s just bad. “Done” in the terminal tells you nothing about whether the screen is right.
That judgment is the part that’s still yours, and you can’t do it from a terminal, because the page you’d be judging never leaves your machine. The result has to travel to where you are, the way the text already does.
FAQ
Why can't I see my dev server on my phone when I run Claude Code remotely?
Because the terminal session is text and text travels over any connection, but the rendered page sits at localhost on your machine. Localhost means that machine, and your phone is a different machine, so the one output you need to judge frontend work is the one that doesn't follow you to the phone.
Why isn't reading the terminal output enough to know the UI is right?
The agent reports that it finished and shows you a diff, but done and a diff don't tell you whether the rendered screen looks right. Agents confidently complete UI tasks and still produce something different from what was asked, so you have to look at the actual page to judge it.