Claude Code over SSH, in Plain English

Your website lives on a server in a data center. That’s where the files sit when something breaks at 11pm.

SSH is how a person reaches that server. Claude Code, Anthropic’s coding tool, runs in a terminal window, so it can run on that server too.¹ None of what follows assumes you have ever opened a terminal window.

SSH is a remote keyboard

SSH stands for Secure Shell. It’s an open protocol maintained through the internet’s standards process, published in this form in 2006.²

You type on your computer and the commands run on a different computer. The connection is encrypted, so nobody in between can read what passes through it.²

One line opens it:

ssh [email protected]

Everything typed after that goes to the server.

Claude runs where the code is

Claude Code reads the files that make up a site, edits them, and runs commands.¹ It’s a program you start from a command line, which is why it works over SSH.¹

The install is the same single line you would run on a laptop, and the same claude command starts it.³

It runs on current versions of macOS, Windows, Ubuntu, Debian, and Alpine Linux, and wants at least 4 GB of memory.³ If you don’t know which of those your server runs, that’s a one-line question for whoever hosts it.

The payoff is that Claude works on the real files, with the same database and configuration your site runs on. No copying anything to a laptop and hoping it behaves the same there.

What it’s allowed to do there

The server running your live site is not a place to experiment, so it matters what Claude can touch there.

Anthropic’s security documentation is direct about where the responsibility sits: “Claude Code only has the permissions you grant it. You’re responsible for reviewing proposed code and commands for safety before approval.”⁴

In Manual mode, Claude Code starts read-only and asks before it edits a file or runs a command that changes the system.⁴ In that mode it can only write inside the folder you started it in, plus anything below that folder.⁴

Start it in the folder that holds the project, not at the very top of the server’s file system where it can reach everything.

Logging in is the awkward step

The first time you run it, Claude Code opens a browser window so you can sign in. That browser is on your own computer, not the server.

Anthropic documents what happens next. The browser may show you a login code instead of signing you straight back in, and you copy that code and paste it into the terminal. The docs list SSH sessions as one of the situations where this happens.⁵

Nothing is broken when that happens.

If your connection drops, the work stops

An SSH session ends when your laptop sleeps or the wifi drops. Anything running inside that session ends with it, mid-task.⁶

Anthropic recommends the standard fix for this: “To keep a session running on a remote machine after you disconnect from SSH, start it inside tmux or screen.”⁶

tmux holds a session open on the server independently of your connection. Its manual says each session “is persistent and will survive accidental disconnection.”⁷

tmux new -s claude
claude

Close the laptop and the work keeps going on the server. Reconnect later with tmux attach -t claude and the session is where you left it.

Use keys, not passwords

An SSH key is a pair of files. One stays on your computer, the matching one sits on the server, and together they replace typing a password.⁸

NIST’s guidance on SSH access management describes the advantage this way: the access a key grants can be determined by inspecting the server it points at.⁸ You can look at a server and see who can get in.

NIST also names the cost of not tracking them. If the deployment of authorized keys isn’t controlled, “users can create unauthorized trust relationships and access that can be exploited.”⁸

Keys keep working until somebody removes them. In practice that means a contractor who finished a project last year may still have a way in, so ask your developer for the list of who has one.

You might not need SSH at all

Three alternatives cover a lot of what people reach for SSH to do.

Remote Control connects a session running on your own machine to your phone or a browser. Claude keeps running locally the entire time, so your files and tools stay on your machine.⁶

This is the one I use every day. My laptop shows up in the Code tab of the Claude app on my phone, with a green dot when it’s online. Tapping it puts me back in the session I left at my desk.⁶

A setting inside Claude Code turns that connection on automatically for every session, which is why the entry is always there.⁶

Claude Code on the web runs in the cloud against a GitHub repository, with nothing to set up on your own machine.⁹

The desktop app’s SSH sessions connect to a remote machine and install Claude Code there for you, with the app as the interface.¹⁰ I wrote about what those sessions do and don’t do when the Linux build shipped.

Which one to use

Claude Code runs wherever you can open a terminal, and SSH is how you open one on a machine you aren’t sitting in front of.

• Plain SSH when the work has to happen on one specific machine, like a client’s server or a machine with a particular database attached to it.
• Remote Control when the files are already on your own computer and you want to reach that session from somewhere else.⁶
• Claude Code on the web when the code lives in a GitHub repository and nothing needs to run on your machine.⁹
• The desktop app’s SSH sessions when you want the remote machine without the terminal.¹⁰

The key list is worth asking about no matter which one you use.