HubSpot drifts. Company descriptions get written three different ways, deal notes pile up half-finished after a quarter, custom-object copy diverges as it passes between teams, and the duplicates multiply until nobody trusts the account view. The built-in assistant stops reading after a few hundred records, and a CSV re-import writes back blind. One customer had given up on it entirely: "We're literally hiring a consultant to help with duplicates inside of HubSpot."
Scratch pulls your HubSpot CRM down as files on your computer. Your AI reads and edits every record, not the first few hundred, about 10x faster than it works over an API because it reads the files directly instead of making a call per record. Every change comes back as a word-level diff next to the original, and nothing writes back until you approve it, per record. Scratch is the companion app to Whalesync: clean the CRM up here, then let Whalesync keep it in sync once the records are right.
What Scratch edits in HubSpot
- Contacts and companies
- Deals, tickets, quotes, and line items
- Notes, tasks, calls, and meetings
- Custom objects and the associations between any of the above
Scratch pulls the full set: contacts, companies, deals, tickets, quotes, line items, products, notes, tasks, calls, meetings, leads, and any custom objects you have defined. The AI rewrites the prose fields and the associations between records.
How it works
- Scratch pulls your HubSpot CRM into files. Contacts, companies, deals, tickets, quotes, line items, products, notes, tasks, calls, meetings, leads, and custom objects all come down to a folder on your laptop, one file per record. Emails are pulled for context only and stay read-only. Workflows, lists, and Marketing Hub assets (blog posts, landing pages, CTAs) are left where they are.
- Your AI edits the fields you point it at. Open the folder in the agent you already use. Try a prompt on a few records, then let it run across the whole object, every record, not a sample. The AI rewrites the prose fields and the associations between records, and it edits the files, never the live CRM.
- You review every diff and write back. In the Scratch desktop app, each changed field shows next to the original, word by word. Approve what ships, and Scratch writes only the records and only the fields you approved back through the HubSpot CRM API. Untouched fields are left exactly as they were.
What teams use it for
- Standardize company descriptions across every account so the firmographics read the same way.
- Tighten and normalize deal notes after a quarter of drift, before a board update or a forecast review.
- Clean up custom-object copy that diverged as it passed between teams.
- Merge duplicate contacts and companies in one reviewed pass, instead of hiring a consultant.
- Normalize the industry and segment fields across the whole book of business.
- Reconcile copy and associations after a migration, so the source of truth matches again.
Why not let AI write straight to HubSpot?
A direct API write or an MCP server hands the AI the publish button straight to your live CRM, and it is the slow path on top: every record is an API call, and on a real book of business it falls over. There is no diff, no review queue, no rollback. One confident bulk pass rewrites every record at once, fires the workflows watching those fields, and skews the reports your team reads on Monday. By the time you catch the bad edit it is already live, and the original is often gone.
Scratch gives the AI the same access, but against a local copy. The write-back step is pulled out and handed to you. The AI can change anything in the copy; only you commit it, you keep the publish button, and Scratch writes back only the fields you changed. Every published record is reversible on its own, so yesterday's run is never a one-way door.
What's safe, and what's locked
Workflows, lists, and Marketing Hub assets (blog posts, landing pages, CTAs) stay put and are never edited. Emails are pulled for context only and stay read-only. 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. Scratch writes back only the fields you changed, so untouched fields are left exactly as they were, and locked or never-touch fields you flag are stripped before write-back. Every published record is reversible per record, and optional Python validators, which the AI can author, flag edits that break a length cap or touch a field you protected, next to the diff. Scratch is SOC 2 compliant.
Questions HubSpot teams ask
Will editing records fire my workflows or skew my reporting?
Edits in the local copy fire nothing and change no report. When you approve a record, Scratch writes it back like any normal update, so a workflow watching that field can fire, the same as if you had edited the record by hand. The difference is that you choose which records write back, instead of one bulk pass tripping every workflow at once and skewing the numbers.
Can it actually clean up the duplicates we were going to hire a consultant for?
Yes. The AI reads every contact and company, not the first few hundred, and proposes the merges and rewrites in the local copy. You review each one as a diff and approve only the ones that are right. That is the job people give up on by hand, done in one reviewed pass instead of a paid engagement.
Will it touch my emails, workflows, or locked fields?
No. Emails are pulled for context only and stay read-only. Workflows, lists, and Marketing Hub assets are left untouched. Scratch writes back only the fields you change, and locked or never-touch fields you flag are stripped before write-back, so they never reach the live CRM even if the AI edits the copy.
Can I undo a run after it wrote back to HubSpot?
Yes. Every written record is reversible from Scratch, per record. Roll back one record or the whole run, and Scratch restores the original for you to publish. You never have to choose between a bulk hygiene pass and a one-way door.
See it on your own HubSpot
Pull a few hundred records to feel the loop, then point the agent at the whole object.