---
title: "Recover a site"
url: https://docs.corefinity.com/guides/migrate/recover-a-site
type: guide
last_updated: 2026-11-13
feature: migrate
related:
  - https://docs.corefinity.com/mcp/tools/corestack_site_target_facts_get.md
  - https://docs.corefinity.com/mcp/tools/corestack_site_rewrite_urls.md
  - https://docs.corefinity.com/mcp/tools/corestack_site_rewrite_urls_restore.md
  - https://docs.corefinity.com/mcp/tools/corestack_site_smoke_test.md
  - https://docs.corefinity.com/mcp/tools/corestack_site_http_check.md
  - https://docs.corefinity.com/mcp/tools/corestack_site_copy_hold_set.md
---
# Recover a site

Have an AI assistant find a site's last good state and take the least destructive way back to it, once you choose.

## Before you start

- An AI assistant is connected to your organisation, with a token that may change the site: see [Scopes](/mcp/scopes).
- A description of what changed or went wrong, up to 2,000 characters.

## Start the recovery

Say what happened and when. The assistant follows the `recover-site` playbook.

```text
Use the Corefinity MCP prompt "recover-site" with site: shop.example.com,
what_happened: the home page has been blank since the plugin update at 14:00.
Follow it step by step and tell me before anything that changes my account.
```

To find the cause first, without changing anything, see [Diagnose a site](/guides/migrate/diagnose-a-site).

## Choose the way back

The assistant matches what happened against the site's deployments, backups and operations. It offers these ways back, least destructive first, and waits for your choice.

| What went wrong                  | The way back                                                                                              | Do it yourself                                                                                      |
| -------------------------------- | --------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
| A deployment broke the code      | Roll back to the previous release. After a site's first Git deployment, that is its files from before it. | [Deploy, stop, retry and roll back](/guides/sites/code-and-deployments/deploy-stop-retry-roll-back) |
| A cache or a worker is stuck     | Run one of the site's platform tools.                                                                     | [Platform tools](/guides/sites/create-and-manage/platform-tools)                                    |
| Data or files are wrong          | Take a backup of the site as it is, then restore an earlier one.                                          | [Restore a backup](/guides/sites/backups/restore-a-snapshot)                                        |
| A web server rule broke requests | Revert the rules to the last good revision.                                                               | [Web server rules](/guides/sites/site-settings/web-server-rules)                                    |

A PHP, web server or Node.js change that the site refused has already been put back: there is nothing to undo.

## What you decide

- Which way back to take.
- Every change, one at a time. A restore to production also asks for the site's domain: see [Confirmations](/mcp/confirmations).
- Whether to skip the backup taken before a restore.

The assistant never deletes anything. After two attempts that did not work, it stops and reports what it tried.

## What you get

- What was done, and the names of any backups it took.
- The result of a check of the site's pages afterwards.
- What is left for you to do.

## Related

- [Diagnose a site](/guides/migrate/diagnose-a-site): Find the cause of a problem.
- [Backups](/guides/sites/backups/snapshots): Take a backup before a change.
- [Confirmations](/mcp/confirmations): Send the typed value a destructive call needs.
