Tailscale exit node for a cloud agent: browse from home, run in the cloud

An exit node routes a device's internet traffic out through another device on your Tailscale network. Put one on a machine at home and a cloud server's browser leaves from your household address, which is what the sites that block datacenter ranges key on.

The short answer

An exit node is a device on your Tailscale network that other devices route their internet traffic through, so the traffic leaves the internet from the exit node’s address. Put one on a machine at home and a cloud server using it browses from your household connection instead of from a datacenter range, which is the one thing the websites that block cloud servers outright actually key on. Three steps: advertise the home device as an exit node (tailscale set --advertise-exit-node), approve it in the admin console, and point the server at it (tailscale set --exit-node=<ip>). The trade is that all of the server’s traffic then rides your home upload bandwidth and your home device must stay on.

Why an agent operator wants one

A cloud server’s browser fails on two kinds of wall. The first is identity, a headless browser announcing itself as headless, and that one is fixed on the server. The second is the address: in our own measurements from a cloud server with a real, headed browser, nine mainstream sites (Reuters, Bloomberg, Expedia, Reddit, Ticketmaster, Zillow, Quora, Tripadvisor, Etsy) still refused the datacenter address where a laptop’s identical browser passed, and Google web search sat behind a CAPTCHA no agent can solve. The full lists are there. Nothing you do on the server changes its address. Routing its traffic out through a device at home does, and that is the fix practitioners arrived at for hosted agents within days of Grok Bot’s launch: “let the Bot use Chrome on a computer at home via Tailscale.”

How exit nodes work

Ordinarily Tailscale only carries traffic between your own devices; the open internet is reached directly from whichever device you are on. An exit node changes that for the devices you point at it: they install default routes (0.0.0.0/0 and ::/0) through the exit node, so, in Tailscale’s words, it routes “all your public internet traffic” the way a traditional VPN would. The exit node itself can be Linux, macOS, Windows, Android, iOS or tvOS; for the always-on job you want a machine that stays plugged in, which in practice means a Raspberry Pi, a small Linux box, or the Mac that already lives under the desk. Tailscale notes that Android exit nodes are “not performant,” and that the macOS and Windows implementations are “new and still undergoing optimization.” Linux is the well-trodden path.

Setting one up

1. On the home device (Linux). Exit nodes forward packets, so IP forwarding has to be on; Tailscale’s docs say plainly that you “must enable IP forwarding” first.

echo 'net.ipv4.ip_forward = 1'          | sudo tee -a /etc/sysctl.d/99-tailscale.conf
echo 'net.ipv6.conf.all.forwarding = 1' | sudo tee -a /etc/sysctl.d/99-tailscale.conf
sudo sysctl -p /etc/sysctl.d/99-tailscale.conf

sudo tailscale set --advertise-exit-node

On a machine that uses firewalld you may also need to allow masquerading. On macOS or Windows the same step is a menu: Exit Node, then Run Exit Node.

2. Approve it. Advertising is a request. An admin opens the Machines page of the admin console, finds the device with the Exit Node badge, and enables “Use as exit node” in its route settings. Until then nothing routes.

3. On the server.

sudo tailscale set --exit-node=<exit-node-ip-or-name>
sudo tailscale set --exit-node-allow-lan-access=true   # optional: keep reaching the server's own LAN
curl -s https://ipinfo.io/json                          # should now show your home address and ISP

The second line matters if the server talks to anything on its own local network (a database on a private address, say); without it that traffic also tries to leave via home. The third is the proof: the server’s public address is now your home’s. Everything the agent’s browser does from that moment arrives at websites from a residential connection.

What it costs you

Bandwidth. All of the server’s traffic, not just the browser’s, now goes home first and out again. Page loads are bounded by your home connection’s upload speed and the round trip to it. For an agent that reads pages, fine; for one that moves large files, noticeable.
Availability. If the home device is off or the home internet is down, the server has no route out at all. Pick a device that never sleeps, and know how to clear the setting (tailscale set --exit-node= with an empty value) from the server’s own console.
Your address is now the agent’s. Sites see your household IP for everything the agent does. That is the feature, and it is also why the agent’s behaviour becomes your reputation with those sites. Rate limits and blocks it earns land on the address your family uses.
It is all-or-nothing per device. Tailscale routes the device’s whole default route through the exit node. If you want only the browser to leave via home and the rest of the server to keep its own address, you are into per-process routing or a second network namespace, which is beyond this page.

The alternatives

A residential proxy service is the rented version of the same idea: you pay per gigabyte for other people’s household addresses, with none of the uptime dependency on your own house and all of the questions about whose addresses they are. Running the browser itself at home and the agent in the cloud is the other split, and it is what Grok Bot users described doing. And for many sites the answer is simply the API you already have credentials for, where a server’s address is expected traffic. Which one fits depends on how many of the nine sites above are actually in your agent’s day; for most agents it is one or two, which is what makes a single exit node at home a proportionate fix.

Tailscale versus Cloudflare Tunnel covers the other direction, reaching a private server rather than leaving from it.