A help center ages in a way nobody schedules time to fix. The product renamed a feature six months ago and half the articles still use the old word. The tone wandered as different people wrote different docs. A brand refresh landed and the screenshots, the headings, and the voice all need a pass. Intercom's editor is built for one article at a time, so a refresh across the whole help center stays a slow, manual slog you keep putting off.
Scratch pulls the whole help center down as files on your computer. Your AI reads and rewrites every article there, not a sample, and about 10x faster than it works over the Intercom API, because it reads the files directly instead of making a call per article. Every change comes back as a word-level diff next to the original. Nothing reaches Intercom until you approve it, per article.
Scratch is the companion app to Whalesync: Whalesync keeps your content in sync across tools, and Scratch is where you wrangle the help-center copy into shape first.
What Scratch edits in Intercom
- Help center article title, body, and description
- Article state (draft / published)
- Article hierarchy (parent collection or section)
- Collection name, description, and icon
- Translated content
How it works
- Scratch pulls your help center into files. Each article comes down to one local file on your laptop. Conversations come down too, for context, but they are read-only.
- Your AI rewrites the fields you point it at. Open the folder in the agent you already use. Try a prompt on a few articles, then let it run across the whole help center. The AI edits the article title, body, and description, sets the article state between draft and published, moves articles in the hierarchy by changing the parent collection or section, edits collection name, description, and icon, and updates translated content. Conversations stay read-only the whole time.
- You review every diff and publish. In the Scratch desktop app, each change shows next to the original, word by word. Validators check length caps and required fields so an article never publishes broken. Approve what ships, and Scratch writes only the articles you approved back through the Intercom API.
What teams use it for
- Run a help-center refresh across every article in one reviewed pass.
- Do a tone pass after a brand update, so the whole help center reads in one voice.
- Push a feature rename across every article that still uses the old word.
- Migrate from another knowledge base and clean the imported copy into shape.
- Restructure the hierarchy by reassigning articles to the right collections and sections.
- Tidy collection names, descriptions, and icons so the help center is easy to navigate.
Why not let AI write straight to Intercom?
An MCP server or a direct Intercom API write hands the AI the publish button straight to the live help center. It is also the slow path: every article is an API call, and on a real help center it falls over. There is no diff, no review queue, no rule layer, no rollback. One confident pass can publish a draft you were not ready to ship, flip an article live with a wrong fact in it, or trip a required-field rule and break the article for every customer reading it. By the time you spot it, it is already public, and the original copy is often gone.
Scratch gives the AI the same full read and write access, but against a local copy. The publish step is pulled out and handed to you. The AI can rewrite anything; only you commit it. And every published article is reversible on its own, so a bad pass is never a one-way door.
What's safe, and what's locked
Conversations are read-only: Scratch pulls them so the AI has context on what customers actually ask, but they never get edited and never write back. Validators enforce length caps and required fields, so an article cannot publish in a broken state. You bring your own AI: Scratch holds no AI credentials and runs no model, so you sign into Claude, Claude Code, Codex, Cursor, Copilot, Cline, or Windsurf the way you already do. Nothing leaves your machine until you publish, every published article is reversible per article, and the whole thing is Git-backed from the first sync. The optional Python validators can be authored by the AI to flag anything else you want guarded, right next to the diff.
Questions Intercom users ask
Can Scratch edit my Intercom conversations?
No. Conversations are pulled for context, so the AI can see what customers actually ask, but they stay read-only. They are never edited and never write back. What Scratch edits is the help center: article title, body, and description, article state, hierarchy, collection details, and translated content.
Will an article ever publish in a broken state?
No. Validators check length caps and required fields before anything writes back, so an article that is missing a required field or exceeds a length cap is flagged in the review queue instead of going live broken. You also approve every article one at a time, so nothing publishes until you have read the diff.
Can it move articles between collections and sections?
Yes. Article hierarchy is editable, so the AI can reassign an article to a different parent collection or section. Like every other change, the move shows in the review queue and writes back through the Intercom API only after you approve it.
Is this faster than editing articles in Intercom one at a time?
Yes, on the work itself. Intercom's editor handles one article at a time; on local files the AI reads and rewrites every article at once and works about 10x faster than over the API, because it reads files directly instead of making a call per article. The exception is write-back, which runs at the Intercom API's own speed for the articles you approve.
See it on your own Intercom
The fastest way to trust it is to watch it run on your own help center. Pull your articles down, point your agent at them, and read the first diff.