Raspberry Pi Connect is regularly described as a replacement for VNC and SSH. It puts a Pi's desktop or shell in a browser tab, with no port forwarding. The small print made me curious. Screen sharing needs a Wayland desktop and a Pi 4 or newer. So I wanted to know whether Connect replaces the old tools or sits beside them. Going by Raspberry Pi's documentation and a RaspberryTips tutorial, Connect is the shortest route to a Pi over the internet. Setup comes down to turning the client on and signing in. Screen sharing still skips X desktops and older boards. On a home network, SSH and wayvnc are already part of Raspberry Pi OS.
What does Connect actually do?
The Raspberry Pi documentation splits Connect into two parts. One is a web service you open in a browser. The other is a client that runs on the Pi. You sign in on the website, pick your device and choose how to connect. The computer you connect from needs only a browser.
What you get depends on the client running on the Pi. There are two of them. The rpi-connect client offers screen sharing and a remote shell. It ships by default on Raspberry Pi OS Desktop and Full. The rpi-connect-lite client offers the remote shell only. It is the default on Raspberry Pi OS Lite. Both need Raspberry Pi OS Bookworm or later, the docs say.
Screen sharing is the fussier of the two. It does not work on systems that use the X window server. It needs a Wayland compositor, and the docs name wayfire and labwc as working. KDE's kwin compositor is not supported. Only one display is shared, the primary one.
Board support narrows it further. Screen sharing runs only on the Pi 4, 400, 5, 500 and 500+, according to a RaspberryTips tutorial. That leaves a Pi 3 or a Zero with the shell through Connect, but not the desktop.
The remote shell reaches much further back. Raspberry Pi announced it on 25 June 2024 with the Connect 1.2 beta. It works with every model, back to the original Pi 1. The same post says the shell copes better than screen sharing on a slow connection.
Connect dropped the beta label with version 2.5 on 9 May 2025. Personal accounts are free for one user. Organisation accounts move to monthly billing after a free trial.
Security is the part I wanted spelled out. Connect's relay servers sit in London and California. Traffic through them is encrypted, and the relays cannot decrypt it. Only your device and your browser hold the keys, the docs say. The data is neither logged nor stored.
How do I set it up, including headless?
On a Pi with a desktop, setup is short because the client comes preinstalled. The Imager walkthrough covers putting Bookworm on a card. These are the steps from the docs, in order:
- Check that the Pi runs Raspberry Pi OS Bookworm or later.
- Check which client is on the card. Desktop and Full images ship
rpi-connect. Lite shipsrpi-connect-lite, which has the shell only. - Start the client:
rpi-connect on - Sign in from the Pi. The command prints a verification URL.
rpi-connect signin - Open that URL in any browser and sign in to your Connect account.
- On the Connect website, find the Pi in your device list. Choose screen sharing or remote shell.
One rule matters more than the rest. The client runs as part of your user account. So it only runs while that user has an active login session. A board with no screen or keyboard usually has nobody logged in. For that case the docs give this fix:
loginctl enable-linger
Linger keeps your user's services running with nobody logged in. Mobzter's Linux commands guide covers the basic terminal commands.
Linger only covers the shell. Screen sharing also needs a desktop session. A headless Pi you want to see as a desktop needs auto login switched on too.
A headless Pi runs with just power and a network cable, and no screen. That is where auth keys come in. An auth key links a Pi to your account without anyone opening a verification URL on it. Each key links one device, the docs say, so every Pi needs its own. A personal account can hold one active key at a time.
Raspberry Pi Imager can write the key while it flashes the card. That key lasts about six hours, RaspberryTips says. So the card should go into the Pi soon after flashing. The tutorial adds two conditions for this headless route. A username and password must be set in Imager. The Pi also needs an internet connection at first boot.
Is it better than VNC, SSH or Tailscale?
On a home network, the older tools are already part of Raspberry Pi OS. Its VNC server is wayvnc, switched on in the Interfaces tab of Control Centre. SSH is disabled by default, the remote access docs point out. Both need a setting changed, while the Connect client comes installed.
The real difference is reach. VNC is limited to a local network, and Connect needs no port forwarding, the RaspberryTips tutorial explains. Reaching SSH or VNC over the internet means opening router ports or adding another network layer. Connect handles that part through its relays.
RealVNC sells its own remote access product, also called Connect. Its free Home subscription ended in June 2024, as AlternativeTo reported. A search summary of RealVNC's help page says its Connect product does not support Wayland on Linux. The help page itself was not available at the time of writing.
Tailscale solves the reach problem another way. Each device gets a stable address in the 100.64.0.0/10 range. Tailscale says that address stays the same wherever the device is. Its Personal plan is free for up to 6 users with unlimited user devices. It is for non-commercial use only. SSH or VNC still runs over it, so the defaults above still apply.
Speed is the gap in this comparison. None of these sources publishes latency or frame rate figures for Connect, wayvnc, RealVNC or Tailscale. The only performance point is Raspberry Pi's note on the shell and slow connections.
What if it does not work?
The docs list a diagnostics command, and it is the first thing to run:
rpi-connect doctor
Is the Connect icon missing from the menu bar, or the Pi missing from your device list? Then check the login session. The client lives inside your user account. With no active session and no linger, it is not running. On a headless board, loginctl enable-linger is the fix the docs give.
A Pi that offers only a shell has a few possible causes in the sources. The board may be older than a Pi 4. The card may carry rpi-connect-lite, which has no screen sharing. The desktop may run X rather than Wayland. Or it may run an unsupported compositor such as kwin. On a headless board, linger without auto login leaves no desktop session to share. With two monitors attached, only the primary one appears in the browser.
The account side can fail too. An auth key written by Imager lasts about six hours. A card left longer than that may need a new key. On organisation accounts, the docs list a state where the trial has ended without billing set up.
The detail that surprised me most sits in the security notes. Connect has no setting to restrict which relays are used. There is no setting to turn relaying off either. So even the simplest route always has the London and California relays in reserve.
Method note: every figure here comes from Raspberry Pi's documentation and news posts, RaspberryTips, AlternativeTo and Tailscale's own pages. None comes from my own tests.
I test how tiny AI models behave on low-cost boards such as the Pi Zero. My reviews focus on speed and memory limits.
2 articles →