Why I Stopped SSH-ing Into My Servers

February 14, 2026·4 min read·North Brook

sshtailscalecliai-agents

I have six machines on my network. Two run AI agents 24/7. One's a build server. One's a NAS. A Mac Mini handles media. And my daily driver laptop ties it all together over Tailscale.

I SSH into these machines constantly. Or I did.

The old workflow looked like this:

tailscale status          # Which machine was it again?
ssh bryce@100.84.32.17   # Hope I have the right key
# Permission denied. Right, different username on this one.
ssh admin@100.84.32.17
# There we go.

Multiply that by a dozen times a day across six machines and it gets old fast.

The breaking point

I started running autonomous AI agents on a couple of remote servers. Claude Code on one, a custom agent framework on another. These things need babysitting — not constant attention, but frequent check-ins. Is the agent stuck? Did it go off the rails? Is it burning through tokens on a loop?

The SSH dance became unbearable. I needed to go from "I wonder what that agent is doing" to "I'm looking at its output" in seconds, not minutes.

What I built

rc is a terminal UI that shows every machine on your Tailscale network in an interactive picker. You arrow to a machine, hit Enter, and you're connected via SSH. That's it. No IPs to remember, no keys to manage, no config files to maintain.

But SSH is just one mode. Hit Shift+Tab and you cycle through:

Terminal — SSH session with auto-managed keys
Screen — macOS Screen Sharing (VNC) with cached passwords
Cursor — Cursor IDE with Remote SSH, pointed at any directory you choose

The Cursor mode changed my workflow completely. I pick a machine, pick the project directory where an agent is working, and Cursor opens right there with full IDE capabilities. I can read the agent's changes, check git history, look at logs — all in my normal editor, on a remote machine, launched in 3 seconds.

How it works under the hood

rc talks to Tailscale to discover your network. It gets the list of machines, their IPs, OS info, and online status. Everything shows up in a keyboard-driven TUI built with Ink (React for terminals).

On first SSH connection to a machine, rc:

Generates an ed25519 key if you don't have one
Runs ssh-copy-id to distribute it to the remote machine
Saves your username for next time

After that, connections are instant. No prompts, no passwords, no friction.

VNC credentials work the same way — enter once, cached forever (locally in ~/.rc/settings.json).

The keyboard-first philosophy

Every action is a keypress. Arrow keys navigate, Enter connects, / searches, Tab cycles machines, Shift+Tab switches modes, Esc goes back. No mouse needed, no menus to click through.

This matters when you're checking on something quickly. You want muscle memory. After a week of using rc, I don't think about it — my fingers know the path to each machine.

Who this is for

If you have multiple machines on Tailscale and you connect to them regularly, rc will save you time. It's particularly useful if you:

Run a homelab and jump between services
Monitor AI agents on remote servers
Develop on remote machines with Cursor
Just hate remembering IP addresses

Try it

curl -fsSL https://remotecontrol.sh/install | bash

Then just type rc.

It requires Tailscale on all machines and SSH enabled on the targets. Screen Sharing and Cursor are optional — modes only appear when they're available.

The code is open source: github.com/north-brook/remote-control

Built with TypeScript, Bun, and Ink. PRs welcome.