The n8n AI Agent node with Claude: model + memory + tools in one drag-and-drop block
TL;DR
n8n's AI Agent node takes input (chat, webhook, schedule), uses a chat model (e.g. Anthropic/Claude) and can attach memory and multiple tools. The agent decides which tool to call. Keep tools few and well described, write a tight system message, log every run.
Minimal structure
| Block | Choose | Note |
|---|---|---|
| Trigger | Chat Trigger / Webhook / Schedule | Where requests come from |
| AI Agent | Tools Agent | The coordinating brain |
| Chat Model | Anthropic Chat Model | Anthropic API key credential |
| Memory | Simple Memory / Postgres Chat Memory | Per-session context |
| Tools | HTTP Request, Google Sheets, Postgres, Code… | Describe each tool clearly |
Example: order lookup assistant
- Chat Trigger receives a support agent's question.
- The agent has tools Postgres: find order by ID and HTTP: check shipping status.
- System message: “Only answer about orders. Never reveal customer phone numbers. If unsure, say so.”
- Answer goes back to chat and is logged to Google Sheets.
Lessons
- One job per tool with a clear name and description — the agent picks tools from descriptions.
- Write tools (delete, send email) should sit behind a human approval step.
- Cap iterations and timeouts; watch Executions for token-burning loops.