

Claude can build most of a HubSpot setup for you: pipelines, properties, views, reports and workflows. What it can’t do is decide what the setup should be. That part is still yours, or ours.
We’re ProfitPad, a HubSpot partner. For the past month, nearly every client build we’ve shipped was done by Claude working inside the client’s portal, from instructions we wrote. It saves us a lot of time: nobody on our team builds by hand anymore. Here’s how we connect it, what it handles, and where it still needs a person.
There are three ways in. They reach the same portal, but they don’t do the same job.
We build through Chrome. It sees what a person sees, so it can set up views, boards and reports that the connector doesn’t cover. It’s also limited the same way a person is: on a seat that can’t save views, Claude can’t either.
Use the connector to read and analyze. Use the browser to build.
| HubSpot connector | Claude in Chrome | |
|---|---|---|
| Workflows, views, dashboards | No. For workflows, Claude drafts the steps and you build them | Yes |
| Properties | Create only, no edits | Create and edit |
| Deleting | Not possible | Anything your seat can delete |
| Bulk changes | 10 records at a time | One screen at a time, slower |
| Several HubSpot accounts | One at a time; switching means reconnecting | Any account you’re signed in to |
| Audit trail | Every change logged to you and the connector | Logged to you, with nothing marking it as Claude |
| Patient data | Sensitive Data blocks engagement data | Keep it out: not covered by Anthropic’s BAA |
The connector is safer and more predictable. The browser does the work the connector can’t reach, which is most of a real build. That’s also why the browser needs a tighter brief and a review before anyone uses what it built.
Sources: HubSpot connector for Claude, HubSpot MCP server, Using Claude in Chrome safely.
Claude does the clicking well. It doesn’t know your business unless you tell it.
It can:
It can’t, or won’t on its own:
A wrong stage built fast is still a wrong stage.
Claude builds what you describe. A vague brief gets a plausible build that doesn’t fit how your team sells. So the time you used to spend clicking, spend on the brief.
What we put in every brief:
A brief we’d actually send:
In the Northwind portal, build a deal pipeline called New Business with five stages: Discovery, Proposal sent, Negotiation, Won, Lost. Use the existing Lead Source property; check for it before creating anything. Make Closed Lost Reason required on Lost. Build a deal view for reps with Amount, Close date and Next step. Don’t touch the Renewals pipeline or any live workflow. When you’re done, open three test deals and confirm each one shows a next step.
Then review what it built before anyone uses it.
Use both. Breeze is better at coverage inside the portal: tasks, lifecycle stages, missing owners. Claude is better when the answer needs context HubSpot doesn’t hold, like proposals and invoices, and it’s the one that runs our free HubSpot audit, with a person reviewing every finding.
We tested them side by side on our own portal →
Let Claude read freely. Make it ask before it writes.
Our rule for clients: in Claude’s settings for the HubSpot connector, set write tools to Needs Approval. HubSpot recommends the same. Claude can be set to always allow. Don’t.
Three things always get a person’s yes first:
In the browser, Claude can delete anything your seat can, and its changes log under your name. Give it the narrowest seat that can do the job, and never point it at a portal with patient data.
Every client build now runs the same way: we write the brief, Claude builds in the client’s portal, and a person reviews it before the team uses it. No manual building. These are the jobs we point it at.
What we don’t hand Claude: lead routing. Who gets a lead should follow a rule your team can read, like a territory sheet. Claude builds the routing, it doesn’t make the call.
Claude can run on a schedule, which covers checks like a weekly audit. It can’t react the moment a record changes, and it isn’t built to move data between systems every hour without fail.
That’s where we use n8n. Claude helps design and write the automation. n8n runs it on a schedule, on a server, with logs when something fails. We run it for ERP syncs and for jobs too big for HubSpot’s own automation, like a nightly attribution run across more than 30,000 deals.
A simple test: if it reports to a person, schedule Claude. If it moves data or reacts to a change, use n8n.
Want to see what Claude finds in your portal? We run the audit and a person reviews every finding. You get the list of what’s wrong with the records behind each item, the quick wins fixed free, and a scoped plan for anything bigger. Request your free HubSpot audit →
Yes, most of it. Claude can build pipelines, properties, views, reports and workflows from a clear brief, through the HubSpot connector or your browser. It can’t decide what your stages, fields and reports should be. A fast build of the wrong process is still the wrong process, so the brief needs someone who knows how your team sells.
Yes, through your browser. Claude in Chrome works in the HubSpot screens as you, so it can build workflows, views and dashboards from a clear brief. The HubSpot connector can’t: for workflows, it drafts the triggers and actions for you to build. Review every workflow before you turn it on.
Not through the HubSpot connector. It creates and updates records, properties and pipelines, but it doesn’t support deletes. It also respects your required fields and pipeline stage rules. Through a browser, Claude can do whatever your seat can, so keep deletes behind a person’s approval.
Yes, if writes need approval. Claude acts with your own HubSpot permissions and connects to one portal at a time. Set the connector’s write tools to Needs Approval so Claude asks before it creates or updates anything. If your account has Sensitive Data turned on, HubSpot blocks engagement data from Claude.
Yes. Set Claude to run the same audit every week and report what changed: new duplicates, deals stuck in a stage, fields nobody fills in. With write approval on, the audit changes nothing in your portal. A person still decides what to fix, because the report shows what changed and someone has to decide why.
Use both. On our own portal, Breeze was better at coverage inside HubSpot: overdue tasks, lifecycle stages, missing owners. Claude was better when the answer needed context HubSpot doesn’t hold, like proposals and invoices. Both found the same top problem, a pipeline that wasn’t in the CRM. Neither can fix missing records.
HubSpot’s MCP server works with any AI tool that supports MCP, including ChatGPT and Gemini. Some tools, like website page editing, are limited to select clients. We build with Claude, so this guide covers Claude. Check HubSpot’s MCP documentation for what your tool supports before you connect it.