A step-by-step guide using n8n and Claude, from a blank canvas to a working triage bot for your shared inbox. A template and security checklist at the end.
April 2026 · Noted by · reads in ~2 minutes
n8n and Claude together form one of the most useful duos for small impact organisations. n8n handles the workflow side: when things are triggered and where the data goes. Claude handles the language side: what does this email mean, and what would be a good reply?
A running n8n instance (self-hosted or in the cloud), a Claude API key and a shared inbox that you can access via IMAP or Gmail. That’s all.
Create an email trigger in the n8n canvas. Choose Gmail or IMAP, depending on what your organisation uses. Interval: every 15 minutes is sufficient for most applications.
Place an HTTP node after the trigger to send the email body to Claude. The prompt is brief: three to five target categories, and a request for a JSON response containing the category and confidence score.
Important: confidence threshold. Anything below 80% goes straight to the team. No automated handling in cases of uncertainty.
For known categories, make a second Claude call with a tone prompt. Result: a draft response that appears in a side panel or is saved as a draft; it is never sent automatically.
Before the workflow goes live: manually run 20 real emails from the last 30 days through the workflow and check the classifications. An error rate of over 10% means: refine the prompt. Only then go live.
A template for the Claude classification prompt is available on request – just get in touch.
45 minutes, one screen shared, an honest assessment. Not a sales pitch, I show you where I see leverage and where I do not.