How to Run Claude Code on VPS: Complete OpenClaw Setup Guide (2026)
Step-by-step guide to running Claude Code on a VPS with OpenClaw (formerly Clawdbot). Connect Claude to WhatsApp, Telegram, Discord from anywhere.
How to Run Claude Code on VPS: Complete OpenClaw Setup Guide (2026)
Want to run Claude Code (Anthropic's coding agent) on a VPS and access it from WhatsApp, Telegram, or Discord? This guide shows you how using OpenClaw — the open-source gateway that bridges AI agents to messaging platforms.
What is OpenClaw?
OpenClaw (formerly known as Clawdbot) is an open-source project that:
- Runs AI coding agents (Claude, GPT, etc.) on any server
- Bridges them to WhatsApp, Telegram, Discord, iMessage, and more
- Provides a web dashboard for management
- Supports persistent sessions and workspaces
- Works 24/7 on a VPS
Think of it as your personal AI assistant that never sleeps — running in the cloud, accessible from your phone.
Prerequisites
Before we start, you'll need:
- VPS with at least 1GB RAM (Oracle Free Tier, Hetzner, or any Linux server)
- Node.js 22+ installed
- Anthropic API key (for Claude) — Get one at console.anthropic.com
- A messaging account to connect (WhatsApp, Telegram bot token, or Discord bot)
Step 1: Install OpenClaw on Your VPS
SSH into your VPS and run the quick installer:
# Quick install (recommended)
curl -fsSL https://openclaw.bot/install.sh | bash
# Or via npm
npm install -g openclaw@latest
Verify the installation:
openclaw --version
Step 2: Run the Onboarding Wizard
OpenClaw includes an interactive setup wizard that handles most configuration:
openclaw onboard --install-daemon
The wizard will ask you to configure:
- Gateway settings — Where to run the server
- Authentication — Your Anthropic API key
- Channels — WhatsApp QR, Telegram token, Discord token
- Daemon installation — Auto-start on boot
Setting Up Anthropic API Key
When prompted for authentication, choose API Key and enter your Anthropic key:
? Select auth method: API Key
? Enter Anthropic API key: sk-ant-api03-xxxxx...
This key is stored securely in ~/.openclaw/openclaw.json.
Step 3: Connect a Messaging Channel
Option A: WhatsApp (Most Popular)
openclaw channels login
This displays a QR code. Scan it with WhatsApp → Settings → Linked Devices.
Option B: Telegram Bot
- Create a bot via @BotFather
- Copy the bot token
- Add it during onboarding or manually in config:
{
"channels": {
"telegram": {
"token": "YOUR_BOT_TOKEN"
}
}
}
Option C: Discord Bot
- Create a bot at Discord Developer Portal
- Enable Message Content Intent
- Add the bot token to config:
{
"channels": {
"discord": {
"token": "YOUR_DISCORD_BOT_TOKEN"
}
}
}
Step 4: Start the Gateway
If you installed the daemon during onboarding, it's already running. Check status:
openclaw gateway status
To start manually:
openclaw gateway --port 18789 --verbose
Access the web dashboard at: http://YOUR_VPS_IP:18789/
Step 5: Secure Your Setup
Add Gateway Authentication
For production, always add a token:
openclaw gateway --bind lan --token YOUR_SECRET_TOKEN
Or in ~/.openclaw/openclaw.json:
{
"gateway": {
"auth": {
"token": "YOUR_SECRET_TOKEN"
}
}
}
Restrict Access (Optional)
Limit who can message your bot:
{
"channels": {
"whatsapp": {
"allowFrom": ["+1234567890", "+0987654321"]
}
}
}
Step 6: Test Your Setup
Send a message to your connected channel:
- WhatsApp: Send a message to yourself (linked device shows it)
- Telegram: DM your bot
- Discord: Message in a channel where your bot is present
You should get a response from Claude!
Verify from Command Line
openclaw message send --target +1234567890 --message "Hello from OpenClaw!"
Advanced Configuration
Custom Workspace
Your AI agent works in a persistent workspace. Default: ~/.openclaw/workspace/
Add custom files here that Claude can access:
mkdir -p ~/.openclaw/workspace
echo "# My Project Notes" > ~/.openclaw/workspace/NOTES.md
Model Selection
Change the default model:
{
"agents": {
"defaults": {
"provider": "anthropic",
"model": "claude-sonnet-4-5"
}
}
}
Available models:
claude-sonnet-4-5(balanced, recommended)claude-opus-4-5(most capable)claude-haiku-4-5(fast, cheaper)
Enable Thinking Mode
For complex tasks, enable extended thinking:
{
"agents": {
"defaults": {
"thinking": "low"
}
}
}
Options: off, low, medium, high
VPS Provider Recommendations
| Provider | Cost | Notes |
|---|---|---|
| Oracle Cloud | Free | Always Free tier, ARM instances |
| Hetzner | ~$4/mo | Great performance, EU servers |
| Railway | $5/mo | One-click deploy, easy setup |
| Fly.io | ~$5/mo | Global edge deployment |
| DigitalOcean | $6/mo | Reliable, good documentation |
For the free option, see the Oracle Cloud guide.
Troubleshooting
"No auth configured"
Run openclaw health — if it shows no auth, re-run the wizard:
openclaw onboard
WhatsApp disconnects
This usually means the session expired. Re-link:
openclaw channels logout whatsapp
openclaw channels login
Gateway won't start
Check if the port is already in use:
lsof -i :18789
Kill the existing process or use a different port.
Permission denied
Make sure you're running as a user with write access to ~/.openclaw/.
Real-World Use Cases
Once running, here's what you can do:
- Code assistance — Ask Claude to write, review, or debug code
- File management — Claude can read/write files in your workspace
- Web research — Enable Brave Search for real-time information
- Automation — Set up cron jobs for periodic tasks
- Multi-device access — Message from any device, anywhere
Example Conversation
You: Create a Python script that scrapes Hacker News top stories
Claude: I'll create that for you. Let me write the script...
[Claude writes hn_scraper.py to your workspace]
Claude: Done! I've created `hn_scraper.py` in your workspace.
It fetches the top 30 stories from Hacker News and saves them
to a JSON file. Run it with: python hn_scraper.py
Conclusion
Running Claude Code on a VPS with OpenClaw gives you:
- ✅ 24/7 availability
- ✅ Access from any messaging app
- ✅ Persistent workspace and sessions
- ✅ Full coding agent capabilities
- ✅ Privacy (your server, your data)
The setup takes about 15 minutes, and you get a personal AI assistant that's always available.
Ready to start? Visit docs.openclaw.ai for the full documentation.
Resources:
Last Updated: February 2026
Related Articles
AI Safety 2026: How Constitutional AI and RLHF Shape Responsible Development
Explore recent AI safety breakthroughs from Anthropic, OpenAI, and DeepMind. Learn how constitutional AI, improved RLHF, and new alignment techniques are making AI systems more reliable and trustworthy.
AI API Economics 2026: Pricing Models, Optimization Strategies & Market Trends
Compare AI API pricing across providers, analyze cost optimization strategies, and explore LLM economics trends for 2026. Practical insights for developers and businesses.
AI Safety Breakthroughs 2026: Constitutional AI to RLHF Advancements
Explore recent AI safety progress from Anthropic, OpenAI, and DeepMind. Learn about constitutional AI, RLHF improvements, and responsible alignment techniques shaping 2026's AI landscape.