Claude Code + n8n

Module 12.1: Claude Code + n8n

Estimated time: ~35 minutes

Prerequisite: Phase 11 (Automation & Headless)

Outcome: After this module, you will understand n8n basics, know how to trigger Claude Code from n8n workflows, and be able to build visual AI automation pipelines.

---

1. WHY — Why This Matters

You’ve learned to automate Claude with GitHub Actions and scripts. But what if you need:

• Non-developers to create AI workflows?
• Visual debugging of complex pipelines?
• Quick prototyping without writing YAML?
• Integration with 400+ services without coding each one?

n8n gives you visual workflow automation. Drag, drop, connect. Claude Code becomes a node in a larger system. Perfect for teams with mixed technical skills or rapid iteration. While GitHub Actions is code-centric, n8n is visual-first — build AI pipelines without touching YAML.

---

2. CONCEPT — Core Ideas

What is n8n?

n8n is an open-source workflow automation platform — think Zapier but self-hosted. Your data stays with you, and you have full control.

Key features:

• Visual editor with drag-and-drop workflow design
• 400+ integrations (Slack, Google Sheets, databases, APIs)
• Triggers: webhooks, schedules, email, database changes
• Can run any CLI command — this is how we connect Claude Code

n8n + Claude Code Architecture

[Trigger] → [n8n Workflow] → [Execute Claude Code] → [Process Output] → [Action]
    │             │                  │                     │              │
  Email      Visual Editor        claude -p            Parse JSON      Slack
  Webhook    Drag & Drop          Headless             Transform       Database
  Schedule                                                             Trello

Key n8n Nodes for Claude Code

Node	Purpose	Use Case
Execute Command	Run claude -p	Core Claude execution
HTTP Request	Call APIs directly	Alternative approach
Code	JavaScript processing	Parse Claude output
IF	Conditional branching	Route based on AI response
Set	Transform data	Prepare prompts

Installation

# Option 1: npm (recommended for development)
$ npm install -g n8n
$ n8n start

# Option 2: Docker (recommended for production)
$ docker run -it --rm -p 5678:5678 n8nio/n8n

Access the editor at http://localhost:5678.

---

3. DEMO — Step by Step

Scenario: Build a workflow where incoming emails are summarized by Claude and sent to Slack.

Step 1: Install and start n8n

$ npm install -g n8n
$ export ANTHROPIC_API_KEY="sk-ant-api03-..."
$ n8n start

Expected output:

n8n ready on port 5678
Editor is now accessible via: http://localhost:5678

Step 2: Create new workflow

1. Open http://localhost:5678
2. Click “New Workflow”
3. Name it “Email Summarizer”

Step 3: Add Webhook trigger (for testing)

Add a Webhook node:

• Method: POST
• Path: /email-summary

This gives you a test URL like http://localhost:5678/webhook/email-summary.

Step 4: Add Execute Command node for Claude

Add an Execute Command node and configure:

{
  "command": "claude",
  "arguments": "-p \"Summarize this email. At the end, write URGENT: YES or URGENT: NO.\n\nSubject: {{ $json.subject }}\n\nBody: {{ $json.body }}\""
}

Step 5: Parse Claude output with Code node

Add a Code node:

const response = $input.first().json.stdout;
const isUrgent = response.includes('URGENT: YES');
const summary = response.replace(/URGENT: (YES|NO)/g, '').trim();

return [{
  json: {
    summary,
    isUrgent,
    originalSubject: $('Webhook').first().json.body.subject
  }
}];

Step 6: Add IF node for routing

Add an IF node:

• Condition: {{ $json.isUrgent }} equals true
• True branch → Urgent Slack channel
• False branch → Regular Slack channel

Step 7: Add Slack nodes

Add two Slack nodes:

True branch (urgent):

{
  "channel": "#urgent",
  "text": "🚨 *Urgent Email*\n\n{{ $json.summary }}"
}

False branch (regular):

{
  "channel": "#email-summaries",
  "text": "📧 *Email Summary*\n\n{{ $json.summary }}"
}

Step 8: Test the workflow

$ curl -X POST http://localhost:5678/webhook/email-summary \
  -H "Content-Type: application/json" \
  -d '{"subject": "Server down!", "body": "Production server is unresponsive since 3am."}'

Expected: Claude summarizes, detects urgency, routes to #urgent channel.

---

4. PRACTICE — Try It Yourself

Exercise 1: Webhook to Claude

Goal: Create a simple webhook that accepts a prompt and returns Claude’s response.

Instructions:

1. Create workflow with Webhook trigger
2. Add Execute Command: claude -p "{{ $json.prompt }}"
3. Add “Respond to Webhook” node returning the output
4. Test with curl

💡 Hint

The “Respond to Webhook” node lets you return data to the HTTP caller. Connect it after the Execute Command node.

✅ Solution

Workflow: Webhook → Execute Command → Respond to Webhook

Execute Command config:

{
  "command": "claude",
  "arguments": "-p \"{{ $json.prompt }}\""
}

Respond to Webhook config:

{
  "respondWith": "json",
  "responseBody": "={{ $json.stdout }}"
}

Test:

curl -X POST http://localhost:5678/webhook/xxx \
  -H "Content-Type: application/json" \
  -d '{"prompt": "What is 2+2?"}'

Exercise 2: Scheduled Daily Analysis

Goal: Every day at 9am, fetch a webpage and have Claude analyze it.

Instructions:

1. Schedule Trigger: daily at 9:00
2. HTTP Request: fetch a news page or API
3. Execute Command: Claude analyzes content
4. Send email with analysis

💡 Hint

Use the Schedule Trigger node with cron expression 0 9 * * *. The HTTP Request node can fetch any URL.

✅ Solution

Workflow: Schedule → HTTP Request → Execute Command → Send Email

Schedule: 0 9 * * * (9am daily)

HTTP Request: https://news.ycombinator.com/

Execute Command:

{
  "command": "claude",
  "arguments": "-p \"Summarize the top 5 stories from this page:\n\n{{ $json.data }}\""
}

Exercise 3: Multi-Step Pipeline

Goal: Chain multiple Claude calls: analyze → suggest → format.

Instructions:

1. Webhook input with raw text
2. Claude #1: Analyze sentiment and key points
3. Claude #2: Suggest actions based on analysis
4. Claude #3: Format as bullet points
5. Return final output

💡 Hint

Each Execute Command node’s output becomes input for the next. Reference previous outputs with $('NodeName').first().json.stdout.

✅ Solution

Workflow: Webhook → Execute Command (Analyze) → Execute Command (Suggest) → Execute Command (Format) → Respond to Webhook

Node 1 - Analyze:

{
  "command": "claude",
  "arguments": "-p \"Analyze this text for sentiment and key points:\n\n{{ $json.text }}\""
}

Node 2 - Suggest (reference Node 1):

{
  "command": "claude",
  "arguments": "-p \"Based on this analysis, suggest 3 actions:\n\n{{ $('Execute Command').first().json.stdout }}\""
}

Node 3 - Format (reference Node 2):

{
  "command": "claude",
  "arguments": "-p \"Format these suggestions as bullet points:\n\n{{ $('Execute Command1').first().json.stdout }}\""
}

Test:

curl -X POST http://localhost:5678/webhook/pipeline \
  -H "Content-Type: application/json" \
  -d '{"text": "Customer complained about slow delivery times."}'

---

5. CHEAT SHEET

Quick Start

npm install -g n8n
export ANTHROPIC_API_KEY="sk-ant-..."
n8n start
# Open http://localhost:5678

Execute Command Node

{
  "command": "claude",
  "arguments": "-p \"{{ $json.prompt }}\""
}

Parse Output (Code Node)

const output = $input.first().json.stdout;
return [{ json: { result: output } }];

Common Triggers

Trigger	Use Case
Webhook	External events, API calls
Schedule	Cron-based, daily/hourly
Email (IMAP)	Incoming emails
Database	Row changes

Data References

{{ $json.field }}                    // Current node
{{ $('NodeName').first().json.field }}  // Other node
{{ $input.first().json.stdout }}     // In Code node

Useful Nodes

Node	Purpose
Set	Prepare/transform data
IF	Conditional routing
Switch	Multi-way routing
Merge	Combine branches
Error Trigger	Handle failures

---

6. PITFALLS — Common Mistakes

❌ Mistake	✅ Correct Approach
Claude CLI not in n8n’s PATH	Use full path: /usr/local/bin/claude
API key not available	Set ANTHROPIC_API_KEY before starting n8n
No timeout on Claude calls	Set timeout in Execute Command (60000ms+)
Quotes breaking prompts	Use Set node to prepare complex prompts
Ignoring Claude errors	Check exitCode and stderr in output
Huge prompts in command line	Write to temp file, pass file path
No error handling	Add Error Trigger node to catch failures

---

7. REAL CASE — Production Story

Scenario: Vietnamese marketing agency receives 50+ client briefs daily via email. Manual processing took 2 hours every morning.

Problem: Briefs arrived in different formats. Staff had to read each one, extract requirements, log to spreadsheet, create project cards, and notify teams. Error-prone and slow.

n8n + Claude Solution:

[Email Trigger] → [Claude: Extract] → [Claude: Suggest] → [Google Sheets]
                                                              ↓
                                          [Slack] ← [Trello: Create Card]

Implementation:

1. Email arrives with client brief
2. Claude extracts: client name, deadline, requirements, budget
3. Claude suggests: team assignment, approach, timeline
4. Auto-logs to Google Sheets
5. Creates Trello card with all details
6. Notifies appropriate Slack channel

Results (after 2 months):

• Processing time: 2 hours → 5 minutes (just review)
• Zero missed briefs
• Consistent data extraction format
• Marketing manager modifies workflow herself — no engineering needed

Quote: “n8n let our marketing manager build AI automation without asking engineering. She just drags and drops. We went from ‘can you build this?’ to ‘I already built it.’”

Cost: n8n is free (self-hosted). Claude API: ~$30/month for 50 briefs/day.

---

Next: Module 12.2: Workflow Patterns →