HubSpot's Claude connector can create and update CRM records but cannot delete them. Set its write tools to Needs Approval first

Chatting with your CRM is useful. Letting a chat window change your pipeline without a check is a different decision.

What happened

HubSpot published a tested guide to using its connector for Claude, and its knowledge base spells out the controls. The connector lets Claude view contacts, companies, deals, tickets and custom objects, plus engagement history like emails, calls, meetings, tasks and notes. It can create and update records, change properties including deal stage, log notes and create tasks. It does not support delete operations.

HubSpot recommends one setting in particular: in Claude's connector tool settings, set Write tools to Needs Approval. If you choose Always allow, it warns, edits may occur without asking for approval.

Other controls worth knowing:

• Every create and update is recorded in HubSpot's Audit Log against both the user and the Claude connector.
• Users only see data their existing HubSpot permissions allow.
• Super Admins approve the connector and choose its data permissions. Adding new objects later means updating permissions, and each user has to reconnect.
• If Sensitive Data is turned on, Claude cannot access engagement data.
• Per HubSpot's guide, the Claude side needs a paid plan: Pro, Max, Team or Enterprise.

My take

Treat this connector like a new hire with write access to the pipeline. Useful on day one, but you would not let them move deals unchecked.

How I would roll it out for a sales team:

1. Write tools on Needs Approval for everyone, no exceptions, for the first month.
2. Grant only the objects the team actually needs. You can add more later.
3. Check the Audit Log for connector changes every week. It tells you who asked for what.
4. List every HubSpot workflow that fires on a deal stage change. A rep typing "mark this Closed Won" in chat can trigger onboarding emails or invoices downstream.

Point 4 is where AI and existing automation collide. The connector is new; the workflows behind it are years old, and nobody remembers all of them.

I set up AI connected CRM work like this alongside the usual jobs: lead routing, pipeline reporting, data cleanup and follow up sequences. See examples at romielwillautomate.dev.