Tutorials
Build 3 AI Automations With n8n and Claude Today
Learn how to build three practical AI automations using n8n and Claude — from email summaries to intelligent agents — without writing a single line of code.
April 4, 2026
What if you could have an AI that automatically summarizes your emails, classifies support tickets, and answers customer questions — and you didn't need to write a single line of Python to make it happen?
That's exactly what we're building today. n8n is a source-available workflow automation platform (think Zapier, but you can self-host it and actually see what's going on), and when you connect it to Claude via the Anthropic API, you get AI-powered automations that are shockingly easy to set up. As of April 4, 2026, Claude Sonnet 4.6 costs just $3 per million input tokens — cheap enough to run on every email, ticket, or document flowing through your business.
In this n8n Claude automation tutorial, we'll build three real workflows from scratch. No prior n8n experience required.
What You'll Build
By the end of this guide, you'll have three working n8n workflows powered by Claude:
- An email summarizer that watches your inbox and sends daily digests to Slack
- A content classifier that auto-tags and routes support requests by urgency
- An AI agent with tool access that looks up data, reasons through problems, and takes action
Each one takes roughly 10 minutes to set up. And yes, they all run automatically once you activate them.
Prerequisites
Before we start, gather these:
- An n8n instance — either n8n Cloud (free trial available) or self-hosted via Docker
- An Anthropic API key — sign up at console.anthropic.com
- Basic API familiarity — you don't need to be a developer, but knowing what JSON looks like helps
If you want to self-host n8n, here's the fastest path:
Bash
docker run -it --rm --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n n8nio/n8n
That spins up n8n at localhost:5678. Done.
Step 1: Configure Your Anthropic Credentials
To connect Claude to n8n, add your Anthropic API key as a credential in n8n's settings — this takes about 30 seconds and makes Claude available across all your workflows.
- Open your n8n instance and go to Settings → Credentials
- Click Add Credential and search for "Anthropic"
- Select Anthropic API from the list
- Paste your API key from console.anthropic.com
- Click Save — n8n tests the connection automatically
If the connection test fails, double-check that your API key has the right permissions and that billing is active on your Anthropic account. A key without billing will authenticate but reject actual API calls.
That credential is now available across all your workflows. You set it up once.
Step 2: Build an Email Summarizer Workflow
This is our warm-up — a classic automation that pays for itself on day one. The flow: incoming emails hit a trigger, Claude reads and summarizes them, and the summary gets posted to Slack.
Create the workflow
- Click New Workflow in n8n
- Add an Email Trigger (IMAP) node and configure it with your email server credentials
- Add a Basic LLM Chain node after the trigger
- Inside the LLM chain, select Anthropic as your chat model and choose your saved credential
- Connect them by dragging from the trigger's output to the chain's input
Configure the Claude model
In the Anthropic chat model settings:
- Model:
claude-sonnet-4-6— Sonnet 4.6 is the sweet spot for summarization tasks. Fast, cheap, and plenty smart. - Prompt template:
Summarize this email in 2-3 bullet points. Focus on action items and deadlines.
From: {{ $json.from }}
Subject: {{ $json.subject }}
Body: {{ $json.text }}
- Max Tokens: 300
- Temperature: 0.2 (we want consistent, factual output — not creativity)
Add the output
- Add a Slack node (or Email Send, or HTTP Request — whatever suits your team)
- Map the chain's output to the Slack message body
- Click Save then Activate
So every time an email lands in your inbox, Claude reads it and drops a summary in Slack. The cost? As of April 4, 2026, a typical email summary runs well under $0.001 at Sonnet 4.6's $3/$15 per million token pricing. You could process 10,000 emails for a few dollars.
Step 3: Create a Content Classification Pipeline
This one's more useful for teams. We'll build a workflow that takes incoming support requests via webhook, classifies them by category and urgency, and routes them to the right people. It's like having a triage nurse for your support queue.
Set up the webhook
- Create a new workflow
- Add a Webhook node as the trigger
- Set the HTTP method to POST
- Copy the webhook URL — this is where you'll send support requests
Add the classification step
- Add a Basic LLM Chain node with an Anthropic model and this prompt:
You are a support ticket classifier. Analyze this ticket and respond with ONLY valid JSON.
Ticket: {{ $json.body.message }}
Customer: {{ $json.body.customer_name }}
Respond in this exact format:
{
"category": "billing|technical|feature_request|bug_report|other",
"urgency": "low|medium|high|critical",
"summary": "one sentence summary",
"suggested_team": "engineering|support|sales|product"
}
Set the temperature to 0. You want deterministic classification, not creative flair.
Route based on classification
-
Add a Code node to parse Claude's JSON response
-
Add a Switch node that routes based on the
urgencyfield:- critical → Send to PagerDuty + Slack #urgent
- high → Post to Slack #support-priority
- medium/low → Create a ticket in Linear, Jira, or whatever you use
-
Add the appropriate output nodes for each branch
Structured JSON output is one of Claude's strongest capabilities. Sonnet 4.6 produces valid JSON well over 99% of the time when you give it a clear schema to follow.
But here's the pro tip: add a validation step right after the LLM node. If Claude returns something unexpected (rare, but it happens), you want a fallback path — not a broken workflow.
Test it
Send a test request to your webhook:
Bash
curl -X POST YOUR_WEBHOOK_URL \
-H "Content-Type: application/json" \
-d '{"customer_name": "Jane Smith", "message": "I was charged twice for my subscription this month and need a refund ASAP"}'
You should see it classified as billing, high urgency, routed to the support team. Pretty satisfying when it works on the first try.
Step 4: Wire Up an AI Agent With Tools
This is where n8n Claude automation gets interesting. n8n has a dedicated AI Agent node that lets Claude use tools — it can query databases, call APIs, and make multi-step decisions autonomously.
What we're building
An agent that handles customer inquiries by:
- Looking up the customer in your database
- Checking their subscription status
- Drafting a personalized response
- Escalating to a human if it can't resolve the issue
Set up the agent
- Create a new workflow with a Webhook trigger
- Add an AI Agent node
- Select Claude Sonnet 4.6 as the chat model (or Opus 4.6 for tasks requiring stronger reasoning — but at $5/$25 per million tokens, save Opus for the hard stuff)
- Write a system prompt:
You are a customer support agent for our SaaS product.
Use the available tools to look up customer information
before responding. Be helpful, concise, and professional.
If you cannot resolve the issue, escalate to a human agent.
Add tools to the agent
This is the powerful part. In the AI Agent node, add tool sub-nodes:
- Customer lookup tool: An HTTP Request node pointing to your customer API. Name it
lookup_customerwith a description like "Look up customer details by email address" - Knowledge base tool: Another HTTP Request node pointing to your docs/FAQ endpoint. Name it
search_knowledge_base - Escalation tool: A Slack node that posts to your support channel. Name it
escalate_to_human
The AI Agent passes these tools to Claude, and Claude decides which to call — and in what order — based on the customer's message.
How it works at runtime
When a customer sends a message, Claude will:
- Call
lookup_customerto pull account details - Call
search_knowledge_baseif it needs product information - Draft a response using that context
- Call
escalate_to_humanonly if it genuinely can't help
And all of this happens in a single workflow execution. No code. No orchestration logic on your end. Claude handles the reasoning; n8n handles the plumbing.
If you prefer building agents in code, check out our guide to building AI agents with LangChain for a Python-based approach.
The AI Agent node is genuinely one of the best low-code agent implementations available right now. It gives you tool-use capabilities that would take hundreds of lines of Python to build from scratch — and you can modify the behavior by editing a text prompt instead of refactoring code.
n8n Claude Automation Pitfalls (and How to Avoid Them)
Here are the mistakes that commonly trip people up with n8n and Claude workflows:
Token limits matter. If you're passing large documents to Claude, you'll either hit context limits or rack up costs. As of April 4, 2026, Sonnet 4.6 supports a 1-million-token context window — but that doesn't mean you should fill it. Trim your inputs. Summarize documents before classifying them.
Set timeouts. Claude API calls typically return in 2-10 seconds, but agent loops with multiple tool calls can take 30+ seconds. Check your node timeout settings and adjust them if workflows are timing out.
Use Sonnet for most tasks. It's tempting to throw Opus 4.6 at everything, but Sonnet handles classification, summarization, and straightforward Q&A just as well — at a fraction of the cost. Save Opus for multi-step reasoning and complex analysis where the quality difference actually shows.
Handle errors. Add an Error Trigger workflow that catches failures and alerts you. API rate limits, malformed inputs, and network timeouts will happen. Plan for them.
Pin your model versions. Use specific model IDs like claude-sonnet-4-6 instead of generic aliases. This prevents your workflows from silently breaking when model versions update.
Testing Your Automations
Before flipping any workflow to production:
- Use n8n's Manual Execution button to test with sample data
- Click on each node to inspect its output individually
- Send edge cases through your webhooks — empty messages, extremely long inputs, non-English text
- Monitor your API usage at console.anthropic.com to verify costs match expectations
- Run each workflow 10-20 times with varied inputs before activating the trigger
n8n keeps execution history for every workflow run, including the full input and output of every node. Use it. It's the fastest way to debug when something goes sideways.
Where to Go From Here
You've got three working AI automations. Here's how to push further:
- Add memory: n8n supports memory nodes that let Claude maintain context across interactions — useful for ongoing support conversations
- Chain workflows: Use the Execute Workflow node to trigger one workflow from another for complex multi-stage pipelines
- Schedule recurring tasks: Swap webhook triggers for Schedule Triggers to run automations on a cron — daily reports, weekly summaries, hourly data pulls
- Connect more data sources: n8n has 1,500+ built-in integrations. Wire Claude into Google Sheets, Notion, Airtable, PostgreSQL, or your own APIs
For more Claude workflows, our Claude Code tutorial covers terminal-based AI development.
The real power of n8n Claude automation isn't any single workflow. It's the combination — dozens of small, focused workflows that collectively handle the grunt work your team shouldn't be doing by hand.
Sources
