How to Run Claude Inside an n8n Workflow (2026)

Lead

You can embed Claude directly into n8n workflows to automate reasoning tasks,tag inbound emails, synthesize data, score leads, or run tool-calling agents. Claude gives n8n's visual automation layer a brain; n8n supplies the connectors, scheduling, and routing. The native Anthropic Chat Model node and AI Agent node make both patterns production-ready.

What you'll do: Create an n8n workflow, add the Anthropic Chat Model or AI Agent node, authenticate with an API key, and configure the model and output schema.

Who this is for: Operations teams and technical leaders at SMBs who want to automate repetitive reasoning work without building custom orchestration.

Verified against:** n8n Cloud, Anthropic Claude API. Current as of 2026-05-04.

Prerequisites

How ScaleIT handles this: See AI Quick-Start

An n8n Cloud account or self-hosted instance (version 2024+)
An Anthropic Console account with Claude API access
An Anthropic API key from https://console.anthropic.com/account/keys
Familiarity with n8n's visual workflow editor

Steps

Log into n8n and create a new workflow

Click Create → New workflow
Click Add a node and search for "Anthropic"

Add the Anthropic Chat Model node

Select Anthropic Chat Model from the results
This node sends a prompt to Claude and returns the response

Create Anthropic credentials

In the node's credential dropdown, click Create new credential
Paste your Anthropic API key into the API Key field
Name the credential (e.g., "Anthropic Production") and save

Select the Claude model

In the node's Model dropdown, choose from:
claude-3-5-sonnet-20241022 (fast, cost-effective; default for most tasks)
claude-3-opus-20250219 (most capable; use for complex synthesis)
claude-3-haiku-20250307 (lightweight; use for high-volume classification)

Write your prompt

In the Prompt field, describe the task clearly
Example: "Classify this email as billing, technical, or sales. Return JSON with category and confidence."
Reference incoming data using n8n syntax: {{ $json.email_body }}

Set output format

Enable JSON mode in Response Format settings if your workflow requires structured output
Define the schema (e.g., { "category": "string", "confidence": "number" })
This ensures downstream nodes receive predictable output

For AI Agent workflows (tool calling)

Select AI Agent node instead if Claude needs to call other n8n nodes as tools
Define Tools by selecting other n8n nodes
Set Max Iterations to 5–8 to prevent runaway loops
Write the System Prompt for the agent's goal

Test and deploy

Click Execute to test against sample data
Verify the output in the Results panel matches your expected schema
Click Activate to enable the workflow on schedule or trigger

Verify

The node's Results panel shows the expected JSON or text output
JSON output contains all required fields with correct types
For AI Agents, the execution trace shows the agent called the correct tools
Test the workflow against 3–5 real examples before production

Troubleshooting

"Invalid API key" error

Confirm you copied the key from https://console.anthropic.com/account/keys
Regenerate the key if it's older than 90 days

JSON output is incomplete or malformed

Rewrite the prompt to be more explicit: "Return valid JSON with no text before or after"
Enable JSON mode in the node's Response Format settings

AI Agent runs out of iterations without answering

Reduce the tool list to 3–6 tools maximum
Rewrite the system prompt to state the goal, not the implementation
Lower Max Iterations and capture partial results for review

Workflow cost is higher than expected

Switch to Haiku for classification work (lower cost, sufficient quality)
Reduce the prompt length; every token costs money
Batch similar requests in one workflow run instead of separate triggers

Next Step

Are you looking to put Claude inside n8n without spending weeks learning the patterns the hard way? We can help you design, build, and scale the right automation for your team. Contact ScaleIt for a free consultation.