Claude Code on a server
Run Claude Code on a server, and let it work while you sleep.
A refactor that takes two hours should not die because you closed the lid. Put Paw on a small Linux server, and your agents keep going. You still talk to them in the same chats, from your Mac or your phone.
What changes
The server does the work. You just look in.
With Paw on a server, the Mac app (or your phone) shows a Paw that runs somewhere else. Everything acts on the server: the agents, the files, the terminals and the browser.
- Agents keep running when your laptop sleeps or goes offline
- A deploy only restarts pawd. A small holder process keeps each agent alive
- Your phone and your Mac see the same chats and the same project
- Paw notifies you when an agent finishes or needs an answer
- Only you can get in: your Tailscale login plus a token, no public port
The short version
Deploy, serve, connect.
The server page has the full steps and the security table. In short:
-
Deploy
One script builds Paw on the server and runs it as a service. You say which login may open it.
scripts/deploy-vps.sh --host user@host --owner you@example.com -
Serve it to your tailnet
On the server, point Tailscale at the owner-only port.
tailscale serve --bg http://127.0.0.1:7291 -
Connect and sign in
In the Paw Mac app, open Settings, then Server, and paste the link. Agents on the server sign in to Claude with a token from
claude setup-token.
Straight talk
When this is worth it, and when it is not.
| Paw on a server | tmux over ssh | |
|---|---|---|
| Agents survive a closed laptop | Yes | Yes |
| A chat per agent, with a status ring | Yes | You watch windows |
| Answer from your phone | Yes, as a chat | A terminal on a small screen |
| Alerts when an agent needs you | Push and Mac notifications | You build it |
| Setup cost | A deploy script, Tailscale, a token | You probably have it |
If you only ever run one agent in one terminal, tmux is enough. Paw earns its place when you run several agents, want to answer from your phone, and want to know without looking which one is stuck.
The real limits
What it needs, and what it does not do.
- You need a small Linux machine you control. Paw runs there as its own user.
- Claude Code has to work on the server. Agents sign in with a token from
claude setup-token, kept in/etc/paw/paw.env. - If the server restarts, Paw comes back with it. Agents start again on their next message.
- Agents that sit idle go to sleep after a day by default, and a starred chat never does. A sleeping agent wakes where it left off.
- Agents run with Claude Code's permission prompts off, so they can run commands on that server. Use a machine and folders you can afford to lose, and keep your work in git.
- The server link carries your token. Don't share it.
- The Paw Mac app is the client for now. The phone uses the browser.
Keep reading
The full server guide
Every step, plus the table of who can reach which port.
Use it from your phone
Tailscale, the home screen, and push notifications.
Try it on your own agents.
Paw is a private beta for Apple Silicon Macs. Write for the next batch and say what you run Claude Code on.