MacAI HK | OpenClaw + n8n Setup Guide

Overview

AI Agent Meets Workflow Automation

OpenClaw is an AI agent that reasons, makes decisions, and takes action. n8n is a visual workflow automation platform with 400+ integrations that executes deterministic tasks. Together, they form a powerful AI automation stack — running entirely on your Mac with local LLMs via Ollama.

🧠

OpenClaw — The Thinker

Intent parsing, decision-making, conversation memory. Powered by local Qwen models via Ollama.

⚡

n8n — The Doer

API calls, data transformations, multi-step automations. 400+ integrations with visual workflow builder.

🔒

100% Local & Private

All AI processing stays on your Mac. No cloud APIs, no data leaving your device, no subscriptions.

Architecture

How the Stack Works

OpenClaw and n8n communicate via webhooks, keeping credentials isolated and workflows inspectable. Ollama runs on the host Mac, providing local LLM inference to OpenClaw.

User

➡

OpenClaw

port 3456

AI Agent (reasoning)

➡

n8n

port 5678

Workflow Engine

➡

External APIs

Slack, Gmail, GitHub...

Ollama

host.docker.internal:11434

Local LLM Runtime (on host Mac)

Prerequisites

Before You Start

Hardware Requirements

Resource	Minimum	Recommended
Apple Silicon	M1 or later	M3/M4 Ultra
RAM	16 GB	96 GB+
Storage	20 GB	50 GB+

Software Requirements

Ollama — installed and running on your Mac (ollama.com)
Docker Desktop (v20.10+) with Docker Compose (v2.0+)
Git
AI Model — pull your preferred model, e.g. ollama pull qwen2.5:32b

No Cloud API Keys Needed

Unlike the original guide which requires Anthropic/OpenAI API keys, this MacAI setup uses Ollama for 100% local inference. Set OLLAMA_API_KEY=ollama-local (any value works — Ollama does not validate API keys).

Setup

Install the Stack

Choose your preferred setup method. Option A is the fastest path; Option B gives you full control.

Recommended

The pre-built stack from github.com/caprihan/openclaw-n8n-stack includes pre-configured Docker Compose, example workflows, and sensible defaults.

1

Clone the Repository

bashgit clone https://github.com/caprihan/openclaw-n8n-stack.git
cd openclaw-n8n-stack

2

Configure Environment Variables

Copy the template and edit for local Ollama use:

bashcp .env.template .env
nano .env

.env# Point to Ollama on host Mac (NOT a cloud API)
OLLAMA_API_BASE=http://host.docker.internal:11434
OLLAMA_API_KEY=ollama-local

# n8n Config
N8N_BASIC_AUTH_USER=admin
N8N_BASIC_AUTH_PASSWORD=your_secure_password_here

# Webhook URLs
N8N_WEBHOOK_BASE=http://n8n:5678
N8N_WEBHOOK_URL=http://localhost:5678

# Timezone
GENERIC_TIMEZONE=Asia/Hong_Kong

3

Start the Stack

bashdocker-compose up -d

4

Verify Services

bashdocker-compose ps

Service	URL	Credentials
n8n	http://localhost:5678	admin / (your password)
OpenClaw	http://localhost:3456	—
Ollama	http://localhost:11434	Running on host Mac

5

Import Pre-Built Workflows

Open n8n at http://localhost:5678, go to Workflows, then Import from File. Select JSON files from the workflows/ directory in the cloned repo.

Manual Setup

Build the stack yourself for full control over every configuration detail.

1

Create Project Directory

bashmkdir openclaw-n8n && cd openclaw-n8n

2

Create docker-compose.yml

docker-compose.ymlversion: "3.8"

services:
  openclaw:
    image: openclaw/openclaw:latest
    container_name: openclaw
    ports:
      - "3456:3456"
    environment:
      - OLLAMA_API_BASE=http://host.docker.internal:11434
      - OLLAMA_API_KEY=${OLLAMA_API_KEY:-ollama-local}
    volumes:
      - openclaw_data:/app/data
    networks:
      - ai-stack
    restart: unless-stopped

  n8n:
    image: n8nio/n8n:latest
    container_name: n8n
    ports:
      - "5678:5678"
    environment:
      - N8N_BASIC_AUTH_ACTIVE=true
      - N8N_BASIC_AUTH_USER=${N8N_BASIC_AUTH_USER:-admin}
      - N8N_BASIC_AUTH_PASSWORD=${N8N_BASIC_AUTH_PASSWORD}
      - N8N_HOST=localhost
      - N8N_PORT=5678
      - N8N_PROTOCOL=http
      - WEBHOOK_URL=${N8N_WEBHOOK_URL:-http://localhost:5678}
      - GENERIC_TIMEZONE=Asia/Hong_Kong
    volumes:
      - n8n_data:/home/node/.n8n
    networks:
      - ai-stack
    restart: unless-stopped

volumes:
  openclaw_data:
  n8n_data:

networks:
  ai-stack:
    driver: bridge

3

Create .env File

.envOLLAMA_API_BASE=http://host.docker.internal:11434
OLLAMA_API_KEY=ollama-local
N8N_BASIC_AUTH_USER=admin
N8N_BASIC_AUTH_PASSWORD=change_me_to_something_secure
N8N_WEBHOOK_URL=http://localhost:5678

4

Start Services

bashdocker-compose up -d

Configuration

Configure OpenClaw for Local LLM

Edit the OpenClaw config to use your local Ollama instance instead of cloud APIs. This is the key difference in the MacAI setup.

OpenClaw Configuration

Mount or edit config/openclaw.json to point to your Ollama models:

openclaw.json{
  "models": {
    "default": "ollama/qwen2.5:32b-openclaw",
    "provider": "ollama",
    "api_base": "http://host.docker.internal:11434"
  },
  "thinking_level": "medium",
  "channels": {
    "default": {
      "webhook_url": "http://n8n:5678/webhook/openclaw-default"
    }
  }
}

Model Selection

The default model ollama/qwen2.5:32b-openclaw requires ~22 GB RAM (Q4_K_M). For Macs with less RAM, try ollama/qwen2.5:14b (~10 GB) or ollama/qwen2.5:7b (~5 GB). Check the Hardware Guide for detailed RAM calculations.

n8n Environment Variables

Variable	Description	Default
N8N_BASIC_AUTH_ACTIVE	Enable basic auth	true
N8N_ENCRYPTION_KEY	Encryption key for credentials	Auto-generated
EXECUTIONS_DATA_PRUNE	Auto-delete old execution data	true
EXECUTIONS_DATA_MAX_AGE	Max age of execution data (hours)	336 (14 days)
GENERIC_TIMEZONE	Timezone for scheduled workflows	Asia/Hong_Kong

Integration

Connecting OpenClaw to n8n

The core integration pattern uses webhooks. OpenClaw sends requests to n8n webhook URLs, and n8n executes the workflow. Both services communicate over the internal Docker network.

1

Create a Webhook Workflow in n8n

In n8n, create a new workflow. Add a Webhook node as the trigger, set HTTP method to POST. Note the webhook URL (e.g. http://n8n:5678/webhook/my-workflow). Add your processing nodes and activate the workflow.

2

Register the Webhook in OpenClaw

Register the n8n webhook as a tool/skill in OpenClaw. The agent will call this URL when it needs to trigger that workflow. Since both services share the Docker network (ai-stack), OpenClaw reaches n8n at http://n8n:5678/webhook/... with no external traffic.

3

Test the Connection

bash# Test from inside the OpenClaw container
docker exec openclaw curl -X POST \
  http://n8n:5678/webhook/my-workflow \
  -H "Content-Type: application/json" \
  -d '{"message": "Hello from OpenClaw!"}'

Community Node (Alternative)

Install the n8n-nodes-openclaw community node for a native integration. In n8n, go to Settings, then Community Nodes, and install n8n-nodes-openclaw. This gives you an OpenClaw node with resource/action dropdowns covering all 20+ built-in tools.

Example Workflows

What You Can Build

Here are practical automation workflows you can build with the OpenClaw + n8n stack, all powered by your local LLM.

📧

Email Triage Agent

Gmail triggers n8n, which sends email content to OpenClaw for classification (urgent / action-needed / informational / spam). Results route to Slack or apply Gmail labels automatically.

🐛

GitHub Issue Triage

GitHub webhook triggers n8n on new issues. OpenClaw analyses type and severity, then n8n applies labels, assigns team members, and posts to Slack.

📊

Document Summariser

Upload documents to a watched folder. n8n detects new files, sends content to OpenClaw for summarisation, and stores results in Notion or a local database.

🔍

AI Fact-Checking

Feed claims to OpenClaw for analysis. The agent uses n8n to query multiple sources, cross-references results, and produces a confidence-scored verdict.

Production Setup

6 Production Workflows

MacAI HK ships six pre-built n8n workflows that power the hkmac.ai storefront. Each workflow connects to real services — Resend for email, Cal.com for bookings, WhatsApp Cloud API for support — and runs on your local Mac via a Cloudflare Tunnel.

Workflow Showcase

See detailed diagrams and node-by-node breakdowns for all 6 workflows on the Production Workflows Showcase page.

Cloudflare Tunnel Setup

Cloudflare Workers (and the public internet) cannot reach localhost. A Cloudflare Tunnel creates a secure, outbound-only connection that routes n8n.hkmac.ai to your local n8n instance on port 5678 — no open ports, no firewall rules.

bash# Install cloudflared
brew install cloudflared

# Create the tunnel
cloudflared tunnel create macai-n8n

# Route DNS
cloudflared tunnel route dns macai-n8n n8n.hkmac.ai

# Run the tunnel (routes n8n.hkmac.ai → localhost:5678)
cloudflared tunnel run macai-n8n

Full Setup Script

A complete tunnel setup script with credential generation and launchd service installation is available at Set Up/03-AI-Source/openclaw-n8n/setup-tunnel.sh.

Environment Variables

Set these variables in the Cloudflare Pages dashboard (Settings → Environment Variables) and in your local n8n .env file.

Variable	Value	Notes
N8N_WEBHOOK_BASE_URL	`https://n8n.hkmac.ai`	Public base URL for all webhook triggers
N8N_WEBHOOK_SECRET	Random 32-char string	Shared secret for webhook authentication
WHATSAPP_TOKEN	WhatsApp Cloud API token	From Meta Business Suite
WHATSAPP_PHONE_NUMBER_ID	Phone number ID	From WhatsApp Business API settings
WHATSAPP_VERIFY_TOKEN	Webhook verify token	Custom string for webhook verification

The 6 Workflows

🛒

WF1: Cart Enquiry Pipeline

Triggered when a customer submits an enquiry from the configure page. Sends a formatted quote email via Resend with full hardware specs and pricing breakdown.

📅

WF2: Post-Booking Onboarding

Triggered by Cal.com booking events. Sends a welcome email with preparation checklist, appointment details, and what to bring to the session.

💬

WF3: WhatsApp Support Bot

Triggered by incoming WhatsApp messages via the Cloud API. Routes queries to OpenClaw for AI-powered responses, with handoff to human support for complex issues.

📋

WF4: Lead Qualification

Triggered by contact form submissions. Scores leads based on budget, use case, and hardware ownership, then routes to the appropriate follow-up sequence.

📦

WF5: Inventory Alerts

Runs on a 6-hour cron schedule and on stock update events. Checks inventory levels in Cloudflare KV and sends restock alerts when units fall below threshold.

🧾

WF6: Invoice Generation

Triggered when a booking is marked as completed. Generates a professional PDF invoice with line items, payment details, and sends it to the client via email.

Import & Activate

1

Open n8n & Import Workflows

Open n8n at http://localhost:5678. Go to Settings → Import Workflow. Import each .json file from the workflows/ directory.

2

Configure Credentials

In n8n, go to Credentials and add your Resend API key. Each workflow that sends email references this credential.

3

Set Webhook URLs

Update the webhook base URL in each workflow to match your N8N_WEBHOOK_BASE_URL environment variable (e.g. https://n8n.hkmac.ai).

4

Activate Each Workflow

Toggle the activation switch in the top-right corner of each workflow editor. Active workflows show a green indicator.

Testing the Workflows

After activating all workflows, verify each one is working correctly:

WF1 (Cart Enquiry) — Submit a test enquiry on the configure page and check your inbox for the quote email.
WF4 (Lead Qualification) — Submit the contact form and verify the lead score is calculated and routed.
WF5 (Inventory Alerts) — Wait for the next 6-hour cron cycle, or trigger the workflow manually from the n8n editor.
WF6 (Invoice Generation) — Mark a test booking as completed in the admin panel and verify the invoice email arrives.

Testing Tip

Use n8n's built-in execution log to inspect each workflow run. Click on any execution to see the data flowing through every node — invaluable for debugging.

Security

Security Best Practices

🔑

Credential Isolation

Store all API keys in n8n's built-in credential store. OpenClaw only knows webhook URLs, never actual credentials. If the agent is compromised, your API keys remain safe.

🛡

Webhook Authentication

Add a shared secret for webhook auth. In n8n, validate x-webhook-secret headers using an IF node after the webhook trigger.

🌐

Network Security

Both services run on an internal Docker network. Only expose the ports you need. Use a reverse proxy with HTTPS for any production deployment.

Webhook Secret Configuration

.env# Add a shared secret for webhook authentication
WEBHOOK_SECRET=your_random_secret_here

In n8n webhook nodes, add an IF node after the webhook to validate: headers.x-webhook-secret === your_secret

Lockable Workflows

Once a workflow is tested and working, lock it so the AI agent cannot modify the integration logic. The agent can only trigger workflows, not change them.

Troubleshooting

Common Issues & Fixes

Services Won't Start

bash# Check for port conflicts
lsof -i :5678
lsof -i :3456

# Check Docker logs
docker-compose logs --tail=50

OpenClaw Can't Reach Ollama

Ensure Ollama is running on the host Mac and the container can reach it via host.docker.internal:11434:

bash# Verify Ollama is running on host
curl http://localhost:11434/api/tags

# Test from inside the OpenClaw container
docker exec openclaw curl http://host.docker.internal:11434/api/tags

n8n Webhook Returns 404

Make sure the workflow is activated (toggle in top-right of workflow editor)
Check that the webhook path matches exactly
Verify you are using the correct HTTP method (GET vs POST)

OpenClaw Can't Reach n8n

bash# Verify both are on the same Docker network
docker network inspect openclaw-n8n_ai-stack

# Test connectivity from OpenClaw container
docker exec openclaw curl http://n8n:5678/healthz

High Memory Usage

Reduce n8n execution history retention via EXECUTIONS_DATA_MAX_AGE
Use a smaller Ollama model if RAM is constrained
Set Docker resource limits in docker-compose.yml

Maintenance Commands

bash# Update services to latest versions
docker-compose pull && docker-compose up -d

# View logs
docker-compose logs -f n8n
docker-compose logs -f openclaw

# Backup n8n data
docker run --rm \
  -v openclaw-n8n_n8n_data:/data \
  -v $(pwd):/backup \
  alpine tar czf /backup/n8n-backup-$(date +%Y%m%d).tar.gz -C /data .

OpenClaw + n8n Setup Guide

AI Agent Meets Workflow Automation

How the Stack Works

Before You Start

Hardware Requirements

Software Requirements

Install the Stack

Clone the Repository

Configure Environment Variables

Start the Stack

Verify Services

Import Pre-Built Workflows

Create Project Directory

Create docker-compose.yml

Create .env File

Start Services

Configure OpenClaw for Local LLM

OpenClaw Configuration

n8n Environment Variables

Connecting OpenClaw to n8n

Create a Webhook Workflow in n8n

Register the Webhook in OpenClaw

Test the Connection

What You Can Build

6 Production Workflows

Cloudflare Tunnel Setup

Environment Variables

The 6 Workflows

Import & Activate

Open n8n & Import Workflows

Configure Credentials

Set Webhook URLs

Activate Each Workflow

Testing the Workflows

Security Best Practices

Webhook Secret Configuration

Common Issues & Fixes

Services Won't Start

OpenClaw Can't Reach Ollama

n8n Webhook Returns 404

OpenClaw Can't Reach n8n

High Memory Usage

Maintenance Commands

Ready to automate with local AI?