Hi all!
Lately I’ve been experimenting more with Claude Code together with the HubSpot MCP and API. What I’d love to build is a workflow where I can talk to Claude about the analysis I want to run, the workflows we’d like to create, or issues my teammates run into, and then have the setup:
- do an initial analysis,
- propose a fix,
- test that fix in our acceptance environment,
- and finally draft a report for me so I can give the final go or add feedback.
Is anyone here running a similar setup and willing to share tips?
One of the main challenges I’m facing is hallucinations: Claude sometimes suggests answers that don’t exist in HubSpot, like filters that aren’t actually available in reports or workflows. I try to ground it by pointing it to the HubSpot Knowledge Base and developer docs, but not all of that information seems to be exposed in a way the model can reliably use.
I’ve seen a few posts from Dharmesh Shah about making HubSpot just as easy for agents as it is for humans, essentially that anything you can do in the UI should also be possible via the API/MCP. I know we’re not fully there yet, but I’m hoping some of you might have practical advice or patterns that work well today.
If you have any questions about my setup, happy to share more details.