How I used Claude AI to extract one page's data from a full database dump and push it back directly via the WordPress MCP — no REST API calls, no manual copy-paste, no downtime.
WP Engine backup dashboard — Download the database dump (.sql file) from the relevant backup checkpoint via the WP Engine user portal.
Claude AI (data extraction) — Paste or upload the SQL dump. Prompt Claude to find the `wp_posts` INSERT row matching the target page by title or slug, and return the `post_content`, `post_title`, and `post_name`.
WordPress MCP — get_pages — Claude calls `get_pages` via the connected MCP to find the current page ID on the live site, matching by slug or title.
WordPress MCP — update_page — Claude calls `update_page` with the recovered `post_title` and `post_content` from the dump. No REST API, no curl, no credential setup.
Verify in WordPress admin — Confirm the page is live, check slug and publish status, and publish if restored as draft.
Recover one broken or deleted page from an existing WP Engine backup — without disrupting the dozens of other pages, posts, and settings on the live site.
WP Engine’s standard restore replaces the entire site with a snapshot. Any content created after the backup date is wiped. There is no built-in ‘restore single page’ option.
‘You need to restore the whole backup to get one page back.’ In reality, the page content is sitting right there in the database dump as a plain SQL INSERT statement. You don’t need to run the full restore — you just need to read the dump, extract the right row, and write it back using the WordPress MCP directly from Claude.
SQL dump is massive and unreadable — A full WP Engine database export is thousands of lines of SQL. Finding the right INSERT for one specific page by hand is impractical.
**Claude reads the dump** and locates the exact row for the target page in the `wp_posts` table — by post ID, slug, or title — and returns just the content you need.
MCP needs to be connected and configured — WordPress MCP only works if it's already set up and authenticated against the right site. Without it, Claude has no write access.
**MCP was already connected** to thesciencetalk.com. Claude called `update_page` directly — no separate auth step, no manual API calls, no credential formatting needed.
Page ID mismatch between backup and live site — If the page was deleted and re-created, its ID in the dump may differ from any existing draft on the live site.
**Claude resolved the ID via MCP** — used `get_pages` to find the current page ID on the live site, then called `update_page` with the recovered content. No guesswork.
Target page fully recovered with original content, slug, and formatting intact
Zero other pages, posts, or settings touched on the live site
No downtime — the live site stayed online throughout the entire recovery
Replicable workflow — same approach works for any page or post in any WP Engine backup
The WordPress MCP made the write-back trivially easy — no auth setup, no curl, just a Claude tool call. The harder part was the SQL extraction. I’d write a short script to pull the target page row out of the dump automatically rather than feeding the whole file to Claude. The MCP approach is now my default for any WordPress content recovery; the days of manual REST API fiddling are behind me.