You have a backlog of jobs you keep avoiding. Refresh the meta description on every one of these 400 posts. Rewrite 200 product descriptions in one consistent tone. Push the new product name across all 1,200 pages. The tools are what make those jobs expensive. The AI built into your platform stops reading after a few hundred records. Half the fields have no bulk upload at all, so the work stays one record at a time. One store owner described the state of the art: "You're literally holding down your mouse and dragging for half an hour."
Scratch downloads your whole catalog as files on your computer. Your AI reads and edits every record there, not the first 200, and about 10x faster than it works over an API or MCP. Every change comes back as a word-level diff next to the original. Nothing reaches Shopify, WordPress, Webflow, or HubSpot until you approve it.
Without Scratch, and with
| the job | without scratch | with scratch |
|---|---|---|
| Rewrite 200 product descriptions in one tone | Open 200 records one by one, or ship one tone problem to 200 live pages. | The AI rewrites all 200 in one pass. You scan the diffs and approve. |
| Bulk-edit 2,500 products | "You're literally holding down your mouse and dragging for half an hour." | Describe the change once. Review the preview. Publish. |
| Refresh meta titles and descriptions on 400 posts | Click 400 records by hand, or trust a blind API write. | Over-length fields flagged in the diff. Approve the rest in one pass. |
| Fill missing alt text on 5,000 images | A job so large you never start it. | The AI drafts every line. You batch-approve what reads right. |
| Apply an SEO audit across 800 pages | The export sits in your downloads folder for a month. | Every finding mapped to a record. You scan the diff and ship. |
| Push a rebrand across 1,200 pages | One wrong replacement is live everywhere, and the original is gone. | Every swap shown next to the original. Reverse any record you change. |
How it works
- Scratch pulls your content into files. Every product, post, and landing page comes down to a folder on your laptop, one file per record. Shopify products, WordPress posts, Webflow collections, Notion docs. Nothing touches the live site.
- Your AI rewrites the fields you point it at. Open the folder in your agent. Try a prompt on one record, then let it run across the whole set. Rewrite every product description in this tone and fix the title case. On local files the AI reads every record and a catalog-wide pass takes minutes, not an afternoon. It edits the files, never the live platform.
- You review every diff and publish. In the Scratch desktop app, each change shows next to the original, word by word. Approve what ships, and Scratch publishes only the records you approved back through each platform's API. What you do not approve never leaves your laptop.
You can edit 2,500 products by describing the change once.
The pattern from real customer calls is the same every time: the job was never hard, it was one-by-one. One store owner asked for "something like a Cursor for Shopify," because product descriptions could only be updated one at a time. Another was bulk-editing 2,500 products by holding down the mouse and dragging; his bar for switching was "type in a prompt and it gives you the preview, then we would be using this tool immediately." That is the loop Scratch runs. You describe the change once, the AI edits every file, the preview is the diff, and you approve what ships.
It is fast because nothing sits between the AI and your content. "When it's all on your computer, it's blazingly fast," about 10x faster than the same AI working over an API, because it reads files directly instead of making a call per record.
What marketers use it for
Most people arrive with a job they have been putting off, because doing it by hand means opening hundreds of records one at a time.
- Rewrite 200 product descriptions in one consistent tone across a Shopify store.
- Bulk-edit 2,500 products from one prompt, with a preview of every change.
- Refresh meta titles and descriptions across 400 posts.
- Fill missing alt text across 5,000 images.
- Apply an SEO audit's findings across 800 pages.
- Push a rebrand or product rename across 1,200 pages in one pass.
Why not let the AI write straight to your CMS?
An MCP server or a direct API write hands the AI the publish button straight to your live store and blog. It is also the slow path: every record is an API call, and on a real catalog it falls over. And there is no diff, no review queue, no rule layer, no rollback. One confident pass rewrites every record and ships it, and by the time you spot the wrong tone or the wrong fact it is already live across every page, with the original often gone.
Scratch gives the AI the same full read and write access, but against a local copy on your laptop. Scratch pulls the publish step out and hands it to you. The AI can change anything, only you can ship it, and every published record is reversible. On live pages with no undo, that is the difference that matters.
What's safe, and what's off-limits
Bring your own AI. Scratch holds no AI credentials and runs no model. You sign into Claude, Claude Code, Codex, Cursor, Copilot, Cline, or Windsurf the way you already do, and the AI reads and edits the local files. By default, nothing leaves your laptop until you publish, and every published change is reversible per record. Optional Python validators flag problems next to the diff, like over-length fields or a locked field the agent should leave alone, so a bad edit is easy to catch before you approve. The result is full access for the AI on a local copy, with review and publish kept in your hands.
Questions marketers ask
I have a backlog of bulk content jobs. Can AI do them without me babysitting every record?
Yes, that is the point. The AI does the rewrite across the whole set, hundreds or thousands of records at once. Your job is review, not retyping. Reviewing 500 changes takes about the same effort as reviewing 50, because you are scanning diffs, not opening records one at a time.
Why not just use the AI built into my platform?
Built-in assistants stop reading after a few hundred records; one customer's "just caps out after a maximum of like 200 records." They also cannot bulk-write most fields, so the job stays one record at a time. On local files your AI reads every record, edits in bulk, and works about 10x faster than over an API. You still review every change before it ships.
If I let AI rewrite my copy, will anything hit the live site before I have read it?
No. The AI only edits local files. Nothing publishes until you have seen the change as a word-level diff and approved it. Scratch then ships only the records you approved, back through each platform's API.
One wrong tone or one wrong fact across every page terrifies me. How do I catch that?
Every change is shown next to the original, word by word, before anything ships. A bad tone or a wrong fact shows up in the diff, so you reject it in the review queue instead of finding it live. One customer used the diff to reject every em dash the AI wrote, "because that looks like AI wrote it." Optional validators can also flag edits that break a length cap or touch a locked field, next to the diff.
My content lives in Shopify, WordPress, Webflow, and HubSpot. Does this cover all of them?
Yes. Scratch pulls from Shopify, WordPress, Webflow, HubSpot, Notion, and more, one record per file. The loop is the same everywhere: pull to files, let the AI edit, review the diff, publish only what you approved.
Reviewing 500 changes sounds worse than doing the work by hand. Is it?
It is not. By hand you open every record, read it, edit it, and save it. With Scratch the AI has already made the edit, so you are scanning a word-level diff and approving. Reviewing 500 changes takes about the same effort as reviewing 50. That is the whole shift.
If a published change is wrong, am I stuck with it?
No. Every published change is reversible per record, even on platforms with no native undo. The original sits next to the rewrite until you decide which one stays.
See it on your own content
The fastest way to trust it is to watch it run on your content. See it run on your store and blog, or download Scratch free and run the first pass yourself.