A plain-English field guide
How to set up Claude Code on your Mac, park an always-on server in the cloud, and drive a whole team of AI coding agents from a single window. The exact setup I use every day.
Your laptop is just the remote control. The work happens on the server, and it keeps happening after you close the lid.
Hey, my friend Mark asked me to write this up for you. He runs his whole operation this way and thought you'd get a lot out of it. It's meant to be read start to finish once, then kept as a reference. Nothing here needs deep technical background; every command is spelled out.
Three moving pieces, and that's the whole trick:
The rest of this guide sets those up in order: first Claude Code on your Mac, then the server, then how to tie it all together so it's one keystroke to boot the whole thing.
Start here even if you never touch a server. This alone is worth it. Two things to install, then you're running.
macOS ships with a terminal app, but iTerm2 is the free upgrade everyone uses: better tabs, colors, search, and split panes. Download it from iterm2.com and drag it to Applications. That's the whole install.
Claude Code runs on Node.js (a common piece of developer plumbing). The cleanest way to get it is Homebrew, the standard Mac package manager. Paste these into iTerm2 one at a time:
# 1. install Homebrew (skip if you already have it) $ /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)" # 2. install Node.js $ brew install node # 3. install Claude Code itself $ npm install -g @anthropic-ai/claude-code
Go into any project folder and type one word:
$ cd ~/my-project $ claude # the first time, it opens a browser to log in with your # Claude subscription, then drops you into a chat. That's it. > build me a simple webpage that shows today's weather
The one mental model that matters
Claude Code isn't a chatbot you copy-paste out of. It's in your folder. It can see your files, change them, run your code, and check whether it worked, all with your permission. You describe the goal; it does the mechanical work.
"Droplet" is just DigitalOcean's cute name for a small cloud server. Any provider works (DigitalOcean, Hetzner, AWS Lightsail). It's a Linux computer you rent by the month and reach over the internet. Here's why bothering with one is a genuine unlock.
My server holds all my projects and runs a small army of Claude agents against them around the clock. I kick off a long build ("screen these 400 land parcels and score them"), close my laptop, and check back later from my phone. The server did the work while I was gone. For anything bigger than a quick edit, that's the whole game.
One honest catch: size it right
A too-small server freezes when you pile many heavy agents onto it (I learned this the hard way on a 1-CPU box with no swap memory). If you're going to run a real fleet, get at least 2 CPUs and 8 GB of memory, and add "swap" (a safety valve for memory) on day one. Budget roughly $12–48/month depending on size. A small starter box is ~$6/mo.
Claude Code on your Mac (Part 1): almost certainly yes, if you build anything. The server + fleet layer is worth it under specific conditions. Here's the honest split so you don't over-build.
Reasonable path: start with Part 1 only. Live on Claude Code locally for a couple of weeks. The day you think "ugh, I wish this had kept running while I was out" is your signal to add the server. Don't build the fleet before you feel the pain it solves.
This is the piece people don't expect. Two small tools, mosh and tmux, turn "log into a server" into "press one key and all my agents are back." Here's how each works, in plain terms.
A tougher replacement for the standard ssh connection. Regular ssh drops the moment your wifi hiccups or your laptop sleeps. mosh survives it: change networks, shut the lid, reopen hours later, and your session is simply there, no reconnecting.
Keeps your work alive on the server even when nothing is connected. It gives you tabs that persist: one tab per project, each running its own agent. Disconnect entirely and every tab keeps running, right where you left it.
Put them together and you get what I call the fleet: a set of persistent tabs on the server (pigeon, dra, outreach, one per project), each with its own Claude agent that auto-resumes its last conversation. I don't rebuild it each morning. I wrap the whole thing in a single shortcut:
$ csf # ↑ my alias. behind the scenes it just runs: # mosh your-user@your-droplet-ip -- tmux attach -t fleet # and instantly I'm looking at all my tabs, every agent # still running exactly where I left it: [ pigeon ] [ dra ] [ outreach ] [ ca-consumer ] [ misc ]
csf, or whatever you name it).csf, and you're right back in.The fleet is just a list you control
Which tabs exist is defined by a plain text file (mine is a one-line-per-project list). Add a project, remove one, rename them: it's editing a text file, then rebuilding the session. Your friendly local Claude Code can set this whole thing up for you; you don't have to hand-write it.
csf-style alias) and a small project list for the fleet. Again: hand this to Claude Code and let it build the scripts.You don't have to do the plumbing yourself
The genuinely nice irony: the AI agent is the sysadmin. Once you have Claude Code running locally, you can tell it "help me set up a DigitalOcean server and a persistent tmux fleet I can reconnect to," and it will hand you each step. That's exactly how this setup gets built and maintained.
your-droplet-ip