Every move, step by step.
Ten common moves, from a plain host switch to pulling one subsite out of a network, and what to do when one stops. Each one covers when to use it, the clicks in the dashboard, the same move in WP-CLI, what happens to the site on the receiving end, and what tends to catch people out.
- 01Move a site to a new hostPull the old site in from the new one.
- 02Push a site to another hostSend this site somewhere else, like staging to live.
- 03Big sites and the final syncCopy early, then catch up only what changed.
- 04Restore a backup zipA zip Disembark made, or another host’s backup.
- 05Pull one subsite out of a networkA subsite or Freighter tenant as its own site.
- 06Bring a site into a networkAs a new or replaced subsite or tenant.
- 07Local developmentPull a live site to your computer, safely.
- 08Undo anythingRoll back to any restore point.
- 09When a move stopsResume it, discard it, or let it keep trying.
- 10Scripts and AI agentsEvery move from a terminal, unattended.
The basics #
A few ideas show up in every move. Read these once and the moves below will make sense.
The Move tab
Everything starts at Tools → Disembark → Move. You pick where the site comes from and where it goes to, and Disembark works out what kind of move that is.
| From | To | What runs |
|---|---|---|
| Another site | This site | A pull. This site downloads the other one. |
| This site | Another site | A push. The other site downloads this one. |
| A backup zip | This site | A restore from the zip. |
| Another site or a zip | A new subsite or tenant | A pull or restore into one site of a network or WP Freighter install. Only on its main site. |
There’s no “another site to another site”: run the move on one of them. Every kind saves a restore point before it changes anything, and rewrites every old address on the way in. A pull or push needs Disembark on both sites. A restore only needs it on the site you restore into.
The connect line #
A site’s connect line is how another site, or the Disembark CLI, reaches it. Find it in Tools → Disembark → Settings, or print it with wp disembark connect-line. It looks like this:
disembark connect https://old-site.com 9fXc…Paste the whole line into the other site’s Move tab and the address and token fill in together. Treat it like a password: whoever holds it can copy the whole site, or replace it. wp disembark token --generate makes a new token, and anything holding the old one stops working.
What to bring #
After a pull or push scans the source, it stops and shows you what’s there before anything is copied: every folder with its size, and every table with its size and row count.
Disembark suggests what a move can usually leave behind: old backup archives and dumps in the site root, page and asset caches, PHP error logs, WooCommerce and Action Scheduler logs, email, redirect and 404 logs, and Wordfence’s firewall files and traffic data. Activity logs, visitor statistics, shopping sessions and image-optimizer originals are listed too, but they come across unless you leave them behind.
It also suggests leaving behind what the old server left in the web root: File Manager’s .tmb and .quarantine folders, and nginx.conf and the conf folder plugins write for the old server.
Leave any folder behind, or bring any table empty: its structure comes across and its rows don’t, so plugins that expect the table keep working. Save your choices as a template (“Woo staging copy”) and pick it next time. In WP-CLI that’s --skip, --skip-rows, --template and --bring-everything.
Restore points #
Before a pull, push, restore or search and replace changes anything, the site that’s about to change saves a restore point. It holds a snapshot of the tables the move will replace and copies of the files it will overwrite or remove.
Rolling back to a point puts the site back as it was then. Restore points made after it are undone first, newest first. You can also put back the database only and leave the files as they are. Do it from the screen at the end of the move, from Backups → Restore points, or with wp disembark rollback. See Undo anything.
Catch-up pulls #
After a pull, the site remembers the size and modification time of every file it brought, and a fingerprint of every table over 4 MB on both sides. The next pull from the same source can copy only what changed.
New and changed files come across. Files that are gone from the source are removed here. Tables that haven’t changed on either side are left alone, and anything Disembark isn’t sure about is copied. See Big sites and the final sync.
Pacing and resuming #
Cheap hosts stop PHP after 20 to 60 seconds and fail some requests at random. Disembark works in short steps (20 seconds, or half of max_execution_time if that’s lower), saves where it got to, and carries on in the next request.
It sends bigger batches after quick answers and smaller ones after slow answers, up to three at once on a fast host. A failed request is sent again. A batch that fails twice is split in half. A host that keeps saying it’s busy gets a pause, from 30 seconds up to 5 minutes. A slow host doesn’t make it give up.
Lose your wifi or close the lid for a minute and the dashboard keeps resending for up to 15 minutes. Close the tab and the Move tab lists the move with Resume: a 248,000-file pull that stopped at file 112,000 picks up at 112,000. Click Notify me on a running move (or pass --notify) to get an email, a Slack or Discord message, or a webhook when it finishes or stops.
In WP-CLI, a step gives up after 20 tries again or 15 minutes of failures in a row, and prints the command to resume. --max-retries and --max-retry-minutes change that, and 0 lifts the limit. See When a move stops.
Quiet mode #
A copy of a live store can email customers, run their scheduled jobs and charge cards. Quiet mode stops that. While it’s on, every email is held (you can read them in Settings), WP-Cron and Action Scheduler run nothing, WooCommerce checkout is closed and its webhooks paused, and calls to payment, marketing, shipping-label and email APIs fail straight away. Search engines are asked to stay away.
A pull or restore turns it on before anything arrives, with the Keep this copy quiet switch. A push leaves it off unless you turn on Keep the destination quiet. It’s tied to the address it was turned on for, so when the copy later moves to another address (going live on the real domain), it stands down by itself. Lift it sooner in Settings or with wp disembark quiet-mode off.
Checks before and after #
Before anything changes, the destination checks that the move can finish: enough disk for the files, the restore point and a second copy of the incoming tables; a database user that can create and drop tables; table names that fit in MySQL’s 64 characters; folders PHP can write to. It also warns about older plugins replacing newer ones, PHP or WordPress going backwards, and backups over 180 days old.
After the move, it checks the result. Every table holds as many rows as the source sent. The home page, the login page, the newest post of each type and an archive of each taxonomy load without a PHP error, a 404 or a redirect off the site. The old address is gone. wp disembark verify runs these checks again.
The Disembark CLI #
The Disembark CLI runs on your own computer, not on the server. It backs a site up to a zip (disembark backup), keeps a local mirror up to date (disembark sync), and restores a backup or mirror into a site (disembark restore, with --rollback=<id> to undo). It’s a PHP archive, so it needs PHP on your computer.
wget https://github.com/DisembarkHost/disembark-cli/releases/latest/download/disembark.phar
chmod +x disembark.phar && sudo mv disembark.phar /usr/local/bin/disembark
disembark connect https://example.com <token> Scripts and agents #
Every wp disembark command runs unattended. --yes skips the confirmation, and --format=json prints one JSON summary on stdout, with progress on stderr. Without --yes and without a terminal, a command refuses instead of waiting on a prompt.
Settings has a ready-made prompt for Claude Code, Cursor or Codex with the site’s connect line in it. See Scripts and AI agents.
Removing Disembark #
Delete Disembark from the Plugins screen and its settings, the site token, its cached update check and its daily clean-up go, on every site of a network.
Two things stay on purpose. wp-content/uploads/disembark can hold backup zips and restore points, so delete that folder by hand when you no longer need them. Quiet mode runs from its own mu-plugin and stays on until you lift it, so run wp disembark quiet-mode off first if the copy should behave normally.
Move a site to a new host #
When to use it. You’re leaving one host for another, and the old site still works. Set up WordPress on the new host, then pull the old site in from there.
In the dashboard
- Install and activate Disembark on both sites (Plugins → Add New → Upload Plugin, with
disembark.zip). - On the old site, open Tools → Disembark → Settings and copy the connect line.
- On the new site, open Tools → Disembark → Move. Pick From: Another site and To: This site.
- Paste the connect line into The source’s connect line.
- Decide on Keep this copy quiet. Leave it on while the new site runs at a temporary address: it stands down by itself once the site moves to its real domain.
- Click Pull into this site and confirm.
- Look over What to bring, leave behind anything you don’t need, and carry on.
- When it arrives, read the checks, then log in with the old site’s username and password.
With WP-CLI
# on the old host: print its connect line
wp disembark connect-line
# on the new host: pull it in
wp disembark pull https://old-site.com <token> --yes
# or paste the whole connect line
wp disembark pull disembark connect https://old-site.com <token> --yes Add --notify=you@example.com to hear when it’s done. If it stops, wp disembark pull --resume=latest carries on from where it got to.
What happens to the destination
- A restore point is saved first, labelled “Before pull from old-site.com”.
- Its files and database are replaced with the old site’s, user accounts included.
- It keeps its own
wp-config.php, its own copy of Disembark and its host setup: the mu-plugins it already has, and itsobject-cache.php,advanced-cache.phpanddb.phpdrop-ins. The old host’s platform mu-plugins are left out (--bring-host-filesbrings them). - Tables under its prefix that the old site doesn’t have are dropped. The restore point keeps them.
- Every spelling of the old address is rewritten (http and https, with and without www, inside serialized data, JSON and page-builder CSS), and so are the old server’s file paths.
- Page caches are emptied, PHP’s opcode cache is reset, and the host’s own cache is asked to purge.
Gotchas
- Afterwards the site’s accounts are the old site’s. The login you made on the new host is gone.
- A firewall such as Wordfence that loads through
auto_prepend_filepoints at a path on the old server. Disembark comments that line out and lists it. Set the firewall up again on the new host. - If the old site hid its login page (WPS Hide Login and similar), the end of the move tells you where it is.
- The checks stop the move before anything changes if the new host is short on disk. It needs room for the files, the restore point and a second copy of the incoming tables.
- Update Disembark on the old site too. An older version still works as a source, but it sends its database in one request and can’t help a later catch-up skip unchanged tables.
Push a site to another host #
When to use it. You’re logged in to the site you want to send. The usual case is a staging site that’s ready to go live, but it works for any site the destination can reach.
In the dashboard
- Install Disembark on both sites.
- On the destination, open Tools → Disembark → Settings and copy its connect line.
- On this site, open Move. Pick From: This site and To: Another site.
- Paste it into The destination’s connect line.
- Keep the destination quiet is off by default. Turn it on when the destination is a staging or dev copy. Leave it off for a live site.
- Click Push to … and confirm.
- Choose what to bring. When it’s delivered, open the destination, or click Undo push.
With WP-CLI
WP-CLI has no push command. Run a pull on the destination instead. It does the same thing.
# on this site (the source)
wp disembark connect-line
# on the destination, over SSH
wp disembark pull https://staging.example.com <token> --yes --skip-quiet-mode --skip-quiet-mode keeps quiet mode off, which is what you want when the destination is the live site. Without it, a pull from another address turns quiet mode on.
What happens to the destination
- The destination saves its own restore point before anything changes.
- Its files and database are replaced with this site’s, user accounts included. Afterwards you log in there with this site’s username and password.
- It keeps its
wp-config.php, its own Disembark and its host setup, just like a pull. - If the destination is the main site of a network or WP Freighter install, it asks Where should it go?: a new subsite or tenant, one to replace, or the main site.
- Undo it from this site (Undo push), from the destination’s Backups tab, or with
wp disembark rollbackon the destination.
Gotchas
- Anything that changed on the destination since you copied it (new orders, comments, form entries) is replaced. Think twice before you push staging over a live store.
- The checks warn when older plugin versions would replace newer ones, which is what happens when an old staging site goes over live.
- The destination downloads from this site, so this site has to be reachable from it. A site on your computer can’t push to a live host. See Local development.
- Your browser drives the destination across sites. If its host blocks cross-site requests, or one site is http and the other https, the push can’t start. Pull from the destination instead.
- Pushing from a public site to one on your own computer or network? The browser asks you to allow local network access first.
Big sites and the final sync #
When to use it. The site is big enough that a full copy takes hours, and you don’t want to lose the orders or comments that come in while it runs. Copy it early, test the copy, then run a short catch-up right before you switch DNS.
- Copy earlyRun a full pull days ahead. The old site stays live the whole time.
- FreezeStop edits on the old site, and close checkout or turn on maintenance mode.
- Catch up, then switchPull again with only what changed, check it, and point DNS at the new host.
In the dashboard
- Pull the site ahead of time, as in Move a site to a new host.
- Test the copy at the new host’s temporary address. Make content changes on the old site, not here.
- On switch-over day, freeze the old site: stop editing, and close checkout or turn on maintenance mode.
- On the new site, open Move, pick From: Another site and To: This site, and paste the same connect line.
- Only copy what changed since the last pull appears, switched on. Leave it on, click Pull, and confirm Catch up.
- Watch the Database stop. It says something like “6 copied · 42 unchanged”.
- Check the result, then switch DNS.
With WP-CLI
# days before: the full copy
wp disembark pull https://old-site.com <token> --yes
# switch-over day, old site frozen: only what changed
wp disembark pull https://old-site.com <token> --catch-up --yes
# changed files only, but every table copied again
wp disembark pull https://old-site.com <token> --catch-up --full-database --yes What happens to the destination
- Files that are new or changed on the source come across. Files the last pull brought that are gone from the source are removed here, after the restore point saves them.
- A table over 4 MB is copied only if its fingerprint changed on the source or on this site since the last pull. Unchanged tables stay exactly as they are.
- The restore point saves only the tables the catch-up replaces, so rolling back leaves the others alone. The address rewrite runs only on what was copied.
- After the import, the skipped tables are fingerprinted on the source once more. Any that changed during the catch-up are copied again (“wp_wc_orders changed during the catch-up, copied again”).
- Anything in doubt is copied, with the reason in the log: smaller tables, views, a table without a primary key, a fingerprint that didn’t finish, one site of a network, or a source on an older Disembark.
Gotchas
- A catch-up only works from the same source as the last pull. With no earlier pull, it copies everything.
- For one site out of a network or WP Freighter install, paste the install’s connect line again. Disembark offers the catch-up and picks the site it came from.
- Edits you make on the new site between the first pull and the catch-up are lost in any table the catch-up copies, and a table that changed here is always copied.
- An older Disembark on the source doesn’t report file times or fingerprints. Files are compared by size, every table is copied, and the run says so.
- Some maintenance-mode plugins also block the REST API that Disembark uses. Make sure the catch-up can still connect before you rely on one.
--catch-upcan’t be combined with--into. A catch-up works on a whole site.- Fingerprints take a little time too: about 4 seconds a side for a 2-million-row table.
Restore a backup zip #
When to use it. You have a backup zip, and you want a site to become it: one Disembark made, or a host’s downloadable backup from Kinsta, WP Engine, cPanel and others. Use it when the old host is gone or can’t be reached, or to put a site back the way it was last month.
In the dashboard
- Install Disembark on the site you’re restoring into. The old site doesn’t need it.
- Open Move. Pick From: A backup zip and To: This site.
- Drop the zip in. Its old address is read from the backup. To rewrite from a different address, open Rewrite from a different address.
- If the zip holds several dumps, or one deep in the tree like a cPanel backup’s
mysql/<db>.sql, Move stops before anything changes and asks Which database? Each dump is listed with its size, table prefix and address, and the best guess is picked. A zip with no dump asks whether to Restore the files only. - Leave Keep this copy quiet on for a staging or dev copy. Turn it off if this will be the live site.
- Click Restore into this site and confirm. Keep the tab open while it uploads; the upload goes in pieces sized to the host’s limits.
- Log in with the accounts from the backup.
With WP-CLI
Upload the zip to the server, then run this in the site’s folder.
wp disembark restore ~/example-com.zip --yes
# a host's backup with several dumps: name one
wp disembark restore ~/cpanel-backup.zip --database=mysql/example_wp.sql --yes
# a zip with no database: the files only
wp disembark restore ~/files.zip --files-only --yes --source-url sets the address to rewrite from. From your own computer, the Disembark CLI restores a backup or a disembark sync mirror instead: unzip it, then run disembark restore <url> <folder>.
What happens to the destination
- A restore point is saved first.
- Its files and database come from the zip, user accounts included.
- Addresses are rewritten from the one recorded in the backup to this site’s. With another host’s zip, the table prefix is remapped from the dump’s.
- It keeps its own
wp-config.php, its own Disembark and its host setup. If the backup came from a multisite network, the run lists thedefine()lines to add towp-config.php. - The database goes into copies beside the live tables, and they’re swapped in only once every statement has worked. A failed import leaves the site as it was.
Gotchas
- A backup zip holds the whole database. After a browser backup, Disembark offers to delete the zip from the server. Say yes once it’s downloaded.
- Backups over 180 days old get a warning before the restore starts.
- A restore never ends as files only by accident. With no dump in the zip, the dashboard asks first, and WP-CLI needs
--files-only. - The Disembark CLI on your computer wants a folder, not a zip, so unzip it first.
wp disembark restoreon the server takes the zip as it is. - The browser backup needs PHP’s ZipArchive on the server. Without it, use
disembark backup. - Quiet mode never travels in a backup, so restoring a quiet copy doesn’t make the destination quiet.
Pull one subsite out of a network #
When to use it. One site of a multisite network, or one tenant of a WP Freighter install, needs to stand on its own. Pull it into a standalone WordPress site, or into a tenant of another WP Freighter install. Archived subsites come out too, and so do sites on old networks that started as WordPress MU.
In the dashboard
- Install Disembark on the network (you need to be a super admin) and on the destination.
- Copy the connect line from Disembark on the network.
- On the destination, open Move. Pick From: Another site and To: This site. To land in a WP Freighter tenant instead, do this on that install’s main site and pick To: A new tenant.
- Paste the connect line. If it holds the subsite’s own address, that site comes. If it holds the main site’s address, pick the site under Which site?
- Choose what to bring, then pull. Bring the network’s super admins (as administrators) is on, with their logins listed. Switch it off to leave out the ones with no role on this site.
With WP-CLI
# by the subsite's own address
wp disembark pull https://shop.network.example.com <token> --yes
# or from the main address, naming the site
wp disembark pull https://network.example.com <token> --site=5 --yes
# without the network's super admins
wp disembark pull https://network.example.com <token> --site=5 --no-super-admins --yes
# into a new WP Freighter tenant: run on that install's main site
wp disembark pull https://network.example.com <token> --site=5 --into=new --domain=shop.example.com --yes --site takes a blog ID, a tenant ID, main or the site’s address. From the main address without it, the pull stops before copying anything and lists the sites with their IDs. A whole install isn’t moved as one: that’s refused before anything changes.
What comes across
- Only the site’s own tables. Its users are picked out of the network’s: its members, the authors of its posts, and the network’s super admins, who arrive as administrators. Roles for other sites are dropped.
- Its uploads are flattened.
uploads/sites/5(orblogs.dir/5/files) moves towp-content/uploads, and Freighter’scontent/5towp-content. Every address for those files is rewritten, mapped domains included, and so are an old WordPress MU network’s/site/files/…addresses. - Plugins and themes come across only if this site uses them. Network-activated plugins become ordinary active plugins, except ones that only work on a network (their header says
Network: true), which stay inactive. WP Freighter is switched off on the copy. - The network’s own tables and domain-mapping tables (Mercator, WordPress MU Domain Mapping, Ultimate Multisite) stay behind, and so do tables that subsites removed long ago left behind.
- The install’s own files in the web root (scripts, other apps,
.htaccess) stay behind unless you tick them, or pass--bring-root. Itswp-config.phpnever comes. - The destination saves a restore point first, as with any pull.
Gotchas
- The network’s super admins arrive as administrators unless you switch that off (
--no-super-admins). They’re administrators of that one site, never super admins. - A plugin that reads its settings from the network’s
wp-config.php(a network SMTP plugin, for one) comes across active but can’t work without them. Add itsdefine()lines to the new site’swp-config.php, or turn it off. - To refresh the copy later, paste the network’s connect line again. Disembark offers a catch-up and picks the same site.
- Tables a plugin shares across the whole network come across empty, since they hold every site’s rows.
--shared-tablesbrings the rows, other sites’ included. - A WP Freighter tenant without its own domain can only be reached through the main site’s address. Use the main site’s connect line and pick the tenant.
- On multisite, Disembark needs a super admin. A subsite administrator can’t open it.
- On a WP Multi Network install, the subsite’s own network is used.
Bring a site into a network #
When to use it. You’re gathering sites into one multisite network or WP Freighter install, or refreshing one subsite with a newer copy. Any site can arrive this way: a standalone site, a backup zip, or one site out of another network. A tenant can become a subsite, and a subsite a tenant.
In the dashboard
- Install Disembark on the network’s main site and on the source. A backup zip works too.
- On the source, copy the connect line.
- On the main site, open Move. Pick From: Another site (or A backup zip) and To: A new subsite (A new tenant on WP Freighter).
- Under Where it lands, pick a new site, or Replace an existing one.
- Click Pull in as a new subsite (or Pull into …) and confirm.
- You can push from the source instead. When the destination is a main site, it asks Where should it go?
With WP-CLI
# a new subsite, at /shop/ or shop.network.example.com
wp disembark pull https://shop.example.com <token> --into=new --slug=shop --yes
# replace subsite 7
wp disembark pull https://shop.example.com <token> --into=7 --yes
# a new WP Freighter tenant on its own domain
wp disembark pull https://shop.example.com <token> --into=new --domain=shop.example.com --yes
# a backup zip on the server, as a new subsite
wp disembark restore ~/shop.zip --into=new --slug=shop --yes Run these on the main site. Without --into, a pull replaces the main site itself, and on a network the network still keeps its users. --slug and --domain default to the source’s.
What happens to the destination
- The new site is created right after the restore point is saved.
- The import touches only that site’s tables and folders (
uploads/sites/<id>, orcontent/<id>on WP Freighter). Statements aimed at anything else are left out and counted. - Plugins and themes are added to the shared folders when they’re missing, and never replace a version other sites run. Core files, mu-plugins and drop-ins stay as they are.
- On a network, the incoming users join the network’s users, matched by email. An account that was already there keeps its own password. The site’s posts, comments and WooCommerce orders are updated to match.
- Network-activated plugins stay network-wide. The address rewrite and the checks cover the new site alone.
- Rolling back removes a new site completely, folders included, and puts a replaced one back.
Gotchas
- Quiet mode stays off, because it would cover every site in the install. Don’t bring in a copy of a live store this way unless it’s fine for it to send email and take payments.
- If the network already runs a plugin, the incoming site runs the network’s version of it, even if it brought a newer one.
- Only the main site can do this. Elsewhere, A new subsite is greyed out.
- A new tenant can’t take a domain the install already uses.
- Views, triggers and stored routines in the incoming database don’t come across: a subsite or tenant takes only its own tables and rows. The move log names what it left out.
- On WP Freighter, the new tenant keeps its own users. Log in to it with your accounts from the source.
Local development #
When to use it. You want to work on a live site on your own computer: try an update, build a feature, or chase a bug with real content. Pull the live site down, keep the copy quiet, and pull again whenever you want fresh content.
In the dashboard
- Make a local WordPress site and install Disembark on it. With Cove, that’s
cove add mysite. - On the live site, copy the connect line from Settings.
- On the local site, open Move. Pick From: Another site and To: This site, and paste it.
- Leave Keep this copy quiet on, and pull.
- Work on the copy. Log in with the live site’s accounts. In Cove,
cove login mysitegives you a one-time login link. - To refresh the content later, pull again with Only copy what changed since the last pull switched on.
- To send the whole site back, download a backup zip from Backups on the local site and restore it on the live one. Push won’t work from your computer (see below).
With WP-CLI
# in the local site's folder
wp disembark pull https://live-site.com <token> --yes
# later: only what changed on the live site
wp disembark pull https://live-site.com <token> --catch-up --yes
# see what quiet mode held back
wp disembark quiet-mode emails What happens to the local copy
- A restore point is saved first, so the local site can be rolled back too.
- Quiet mode goes on before anything arrives: emails held, scheduled jobs paused, checkout closed, payment and marketing APIs blocked.
- Login captchas whose keys only work on the live domain (Simple Cloudflare Turnstile, reCAPTCHA and hCaptcha plugins, Wordfence’s login reCAPTCHA) are turned off on the copy, so you can log in.
- A database from a MySQL 8 host is converted where MariaDB differs:
utf8mb4_0900_ai_cibecomesutf8mb4_unicode_520_ci, and the run lists each conversion. - Force-SSL plugins are turned off when the copy runs on plain http. Cove serves every site over https, so this doesn’t come up there.
Gotchas
- You can’t push from your computer to a live host. The destination downloads from the source, and a live server can’t reach
mysite.localhost. Send the site back with a backup zip instead. - Sending the whole site back replaces the live database. Orders, comments and sign-ups since your pull are lost. For code changes, deploy the theme or plugin the way you normally do.
- Pulling between two sites on your own computer? Define
DISEMBARK_DEV_MODEastruein the destination’swp-config.php. Local addresses and self-signed certificates are refused otherwise. - Quiet mode is an mu-plugin, so it keeps working even if you deactivate Disembark. Lift it with
wp disembark quiet-mode offwhen you want the copy to behave like the real thing. - Quiet mode never travels in a backup, so a zip of your quiet local copy doesn’t make the live site quiet.
Undo anything #
When to use it. A move didn’t go the way you wanted, or you changed your mind. Every pull, push, restore and search and replace saved a restore point before it changed anything. Roll back to it.
In the dashboard
- Right after a move, the last screen has an undo button: Undo pull, Undo push or Undo restore.
- Later on, open Backups → Restore points. Each one shows what it was for, when it was made, its size, whether it’s been rolled back, and whether its files were put back already.
- Click Roll back on the one you want. The dialog sums up what will change: the date the database goes back to, how many files the moves placed will go (uploads counted on their own), and how many come back.
- If restore points were made after it, the dialog lists them, newest first. They’re undone first, so the site ends up as it was then.
- Pick Database and files or Database only, and confirm.
- If it undid a pull, push or restore, log in again with the accounts from before.
- A move that stopped partway shows on the Move tab with Resume and Roll back and discard. Cancel during a move rolls back too, once the restore point exists. See When a move stops.
With WP-CLI
# list restore points, newest first
wp disembark restore-points
# roll back to one: later points are undone first
wp disembark rollback 62bcb2c4539c63e4d8c82629 --yes
# the database only: the files stay as they are
wp disembark rollback 62bcb2c4539c63e4d8c82629 --database-only --yes
# undo that one point alone, leaving later moves in place
wp disembark rollback 62bcb2c4539c63e4d8c82629 --only --yes
# the newest one that hasn't been rolled back
wp disembark rollback latest --yes
# free the disk space once you're sure
wp disembark delete-restore-point 62bcb2c4539c63e4d8c82629 --yes
# undo a Disembark CLI restore, from your computer
disembark restore https://new-site.com --rollback=62bcb2c4539c63e4d8c82629 wp disembark rollback prints the later restore points it will undo before it asks you to confirm. wp disembark restore-points shows, for each point, whether its files are kept or were put back.
What a rollback does
- Before it touches anything, it checks the whole snapshot: files present, sizes unchanged, every table there. If anything’s off, nothing changes.
- Restore points made after this one are undone first, newest first, so the site is as it was when this one was saved. Files those later moves placed go, and files they replaced come back.
- The tables the move replaced are put back, one at a time. Files the move overwrote or removed come back, and files and folders it added are removed. With Database only, the files stay as they are and the point keeps its saved copies for a later rollback with files.
- On a standalone site, visitors, wp-admin and WP-Cron get WordPress’s “Briefly unavailable for scheduled maintenance” page while it runs, so nothing loads the site half put back. The rollback’s own steps and WP-CLI carry on. If a rollback stops for good, the page lifts by itself within 10 minutes.
- Plugins and settings the move changed (a captcha or force-SSL plugin it turned off) are put back, and quiet mode the move turned on is lifted.
- A new subsite or tenant is removed completely. A replaced one comes back.
- It’s quick: a 2-million-row table comes back in about 30 seconds.
Gotchas
- Anything changed after the move is lost, such as orders taken on the new site after it went live.
- A rollback moves a point’s saved files back into place, so they go back once. After that the point shows files put back, and rolling back to it again puts back its database only.
- Database only leaves an older database with newer plugin files. That usually works, though a plugin may ask to update its database.
- On a network, each site lists only its own restore points. A point another site made is rolled back from that site, or with
--url=in WP-CLI. Networks and WP Freighter installs don’t get the maintenance page, since it would take down every site there. - A restore point lives on the same server as the site. It’s for undoing a move, not a backup. Keep a backup zip somewhere else too.
- Restore points take disk space. Delete old ones from Backups, or with
wp disembark delete-restore-point all. Settings → Clean up working files keeps them unless you tick Also delete restore points. - To undo a push, roll back on the destination, or click Undo push on the site you pushed from.
- After you roll back a catch-up, the next catch-up treats the rolled-back files as changed and copies them again.
When a move stops #
When to use it. A move stopped partway: the tab closed, the connection dropped, the host refused one request too many, or a plugin on the incoming site crashed. Nothing is lost. A move keeps its progress and picks up where it stopped.
In the dashboard
- Open Tools → Disembark → Move. A move that stopped is listed at the top, with when it started, when it was last active and how far it got.
- Click Resume. The steps that finished are replayed in seconds, and it carries on from the first one that didn’t.
- Or click Discard to throw it away. Once the move got past its restore point, the button is Roll back and discard: the site goes back as it was first.
- A move that stopped after the database arrived but before the old addresses were rewritten still opens here. The site keeps its own address until the rewrite is done, so wp-admin doesn’t send you to the old site.
- Click Notify me on a running move to get an email, a Slack or Discord message, or a webhook when it finishes or stops.
With WP-CLI
# moves that stopped
wp disembark unfinished
# resume one started in the terminal
wp disembark pull --resume=latest --yes
# keep trying a struggling source for up to an hour
wp disembark pull https://old-site.com <token> --max-retry-minutes=60 --yes
# never give up, the way the dashboard does
wp disembark pull https://old-site.com <token> --max-retries=0 --max-retry-minutes=0 --yes
# throw one away, rolling back first if it got that far
wp disembark discard <id> --yes In WP-CLI, a step gives up after 20 tries again or 15 minutes of failures in a row (429, 502, 503 or 504 answers, or a dropped connection), and prints the command to resume. The count runs across the whole run of failures. restore, rollback, discard and search-replace take the same two options.
What resuming does
- Steps that finished are replayed from the record the server kept, not run again. Files aren’t downloaded twice, and the database isn’t imported twice.
- Copying carries on at the file it stopped at, at the pace the host managed before.
- A move that never reached its restore point changed nothing on this site. One that did may be partly changed, and stays listed until you resume, discard or roll it back. Stopped moves with no restore point are cleaned up after a week.
- The incoming site’s roles and logins are renamed to this site’s table prefix in the same request that brings its database in, so a plugin that needs them can’t crash the next step.
Gotchas
- A move started in the dashboard resumes in the dashboard, and one started in WP-CLI resumes in WP-CLI. Either kind can be discarded from both.
- A source that keeps answering busy (429 or 503) gets longer pauses, from 30 seconds up to 5 minutes. If it’s still struggling, resume later, when the host is quieter.
- The checks before a move stop it before anything changes when it can’t finish: short on disk, a database user that can’t create tables, table names too long. Fix the cause and start again, or pass
--skip-checksto go ahead anyway. - If a move keeps stopping at the same step, open Show details on the move for its log, and ask on GitHub with it.
Scripts and AI agents #
When to use it. You want a move to run without anyone clicking: from a deploy script, a scheduled job, or an AI agent like Claude Code, Cursor or Codex. Everything the dashboard does to move a site has a command, and every command runs unattended.
wp disembark connect-line- Print this site’s connect line.
wp disembark token --generate- Make a new token. Anything holding the old one stops working.
wp disembark pull <url> <token>- Pull another site into this one. Takes
--catch-up,--site,--into,--skip,--skip-quiet-mode,--max-retriesand more. wp disembark restore <zip>- Restore a backup zip that is on the server. Takes
--database,--files-only,--intoand--notify. wp disembark unfinished- List moves that stopped.
wp disembark pull --resume=latest- Resume a pull started in the terminal (or pass its id).
wp disembark discard <id>- Throw a stopped move away, rolling back first if it got that far.
wp disembark verify [<id>]- Run the checks after a move again.
wp disembark restore-points- List restore points.
wp disembark rollback <id|latest>- Roll back to one. Later points are undone first.
--database-onlyleaves the files as they are, and--onlyundoes that point alone. wp disembark delete-restore-point <id|all>- Delete restore points.
wp disembark search-replace <from> <to> --dry-run- Preview a rewrite across the database. Without
--dry-runit saves a restore point, then replaces. wp disembark quiet-mode [on|off|emails]- Check, change or read quiet mode.
disembark connect <url> <token>- Remember a site.
disembark backup <url>- Download a backup zip.
disembark sync <url> [<folder>]- Create or update a local mirror.
disembark restore <url> <folder>- Restore a backup or mirror into a site.
disembark restore <url> --rollback=<id>- Undo a CLI restore.
Handing it to an agent
- On the source, open Settings and copy the ready-made prompt for AI agents. It has the connect line in it and links to the agent instructions.
- Paste it into an agent that can reach the destination over SSH, with WP-CLI there.
- Tell it plainly which site is the source and which is the destination. A pull replaces the destination.
- Have it pass
--yes --format=jsonand keep theundoline from the result.
With WP-CLI
wp disembark pull "disembark connect https://old-site.com <token>" --yes --format=json > move.json
jq -r .undo move.json
# a pull that stopped
wp disembark unfinished --format=json
wp disembark pull --resume=latest --yes --format=json The summary includes undo, checks, verify, left_out, caches_cleared and login_url. When a move fails, it has error and step, plus undo or resume when they apply.
What happens on the site
- The same as in the dashboard:
wp disembark pullruns the same steps, saves the same restore point and paces itself the same way. - The checks before the move stop a pull that can’t finish, and list the reasons under
checks. Run it again with--skip-checksto go ahead anyway. - The checks after the move run unless you pass
--skip-verify, and land underverify.
Gotchas
- Without
--yesand without a terminal, a command refuses instead of waiting. Agents should always pass it, after a person has confirmed the move. - A pull started in the dashboard can’t be resumed from the terminal. Resume it from the Move tab, or throw it away with
wp disembark discard <id>. - There’s no
wp disembark push. To push from a script, run the pull on the destination. - The token in a connect line works like a password. Keep it out of commit messages, shared logs and issue trackers.
- Progress goes to stderr. Read only stdout for the JSON.
- A move gives up after 20 tries again or 15 minutes of failures in a row, and the JSON has
resume. For a run nobody watches, pass--max-retries=0 --max-retry-minutes=0to keep trying, or resume it later.
Something not covered here?
Ask on GitHub, or hand the agent instructions to your AI agent.