SSH MCP

SSH MCP does exactly what it says: it lets Claude run shell commands on remote Linux or Windows boxes over SSH. You configure a connection with either password or key-based auth, then Claude can execute commands, read logs, restart services, or poke around file systems without you copying and pasting terminal output back and forth. I've used it mostly for server ops tasks where I'm debugging a deployment or checking service status across a few boxes. Instead of SSHing in myself, I ask Claude to check disk space, tail logs, or restart a service. It's particularly useful when you're troubleshooting something unfamiliar and want Claude to suggest commands and run them in the same conversation. The workflow is: describe the problem, Claude suggests a diagnostic command, it runs it via SSH, you both see the output, iterate from there. Saves a lot of context switching. The setup is straightforward if you're comfortable with SSH keys. You point it at a host, provide credentials, and it connects. Password auth works but key-based is cleaner for anything you'll use repeatedly. Windows support is there via OpenSSH, which is less common but does work if the remote box is configured for it. Quirks: it's a shell command executor, so it inherits all the risks of giving an AI SSH access. You need to trust the host and the commands Claude suggests. There's no sandboxing beyond what you configure on the remote system. Also, it's not a file transfer tool. If you need to move files around, you'll still reach for scp or rsync separately. And because it's community-maintained, expect occasional rough edges if you're running exotic setups. Who shouldn't bother: if you're managing production infrastructure, this is probably too casual. You want proper orchestration tools and audit logs. But for dev environments, staging boxes, or personal servers where you're already SSHing in manually, it's a solid time-saver. I'd also skip it if you're not comfortable reviewing the commands Claude wants to run before they execute.