5,218,750 AI resources indexedWorking toward the world’s largest AI resource index.

PDF

Firewall Rules

An MCP server that provides on-demand skill loading for AI coding assistants.

Installable skillMITBy modbenderUpdated Oct 30, 2026
INTENDED FOROpenClawInstall in a listed agent.

instructions.md

---
name: firewall-14210ff69819
description: Configure firewalls on servers and cloud providers with security best practices.
metadata: {"clawdbot":{"emoji":"πŸ›‘οΈ","os":["linux","darwin","win32"]}}
---

# Firewall Rules

## Critical First Steps
- Allow SSH/remote access before enabling any firewall β€” enabling first locks you out
- Test access in a second session before closing the first β€” verify the rule actually works
- Know how to access provider console β€” it's the only way back if locked out

## Default Stance
- Default deny all incoming traffic β€” only open what you explicitly need
- Default allow outgoing traffic β€” most apps need to reach the internet
- Every open port is attack surface β€” question each one before adding

## Essential Ports
- SSH (22 or custom): Always needed for remote access β€” consider limiting to your IP only
- HTTP (80): Only if serving web traffic β€” also needed for Let's Encrypt HTTP challenge
- HTTPS (443): For production web services
- Don't open database ports (3306, 5432, 27017) to the internet β€” access via SSH tunnel or private network

## Provider Firewalls (Hetzner, DigitalOcean, AWS, etc.)
- Provider firewall applies before traffic reaches your server β€” faster, less server load
- Changes usually apply immediately β€” no reload command needed
- Stateful by default β€” allow inbound, responses automatically allowed outbound
- Apply to server groups for consistency β€” easier than per-server rules
- Provider firewall + OS firewall = defense in depth β€” use both when possible

## IP Restrictions
- Limit SSH to known IPs when possible β€” dramatically reduces attack surface
- Your home IP may change β€” use a VPN with static IP or update rules when it changes
- Allow IP ranges with CIDR notation β€” /32 is single IP, /24 is 256 IPs
- Some providers support dynamic DNS in rules β€” check before building complex solutions

## Common Services to Consider
- VPN (WireGuard: 51820/UDP, OpenVPN: 1194) β€” allows secure access without exposing other ports
- Mail (25, 465, 587) β€” only if running mail server
- DNS (53 TCP/UDP) β€” only if running DNS server
- Monitoring agents may need outbound access to specific IPs

## Docker Warning
- Docker bypasses most OS firewalls by default β€” containers expose ports regardless of UFW/iptables
- Solution: bind containers to localhost only and use reverse proxy for public access
- Or configure Docker to respect firewall rules β€” requires additional setup
- Provider-level firewalls still work β€” they block before traffic reaches Docker

## IPv6
- Firewalls often have separate IPv4 and IPv6 rules β€” configure both
- Provider firewalls may handle both together β€” check their documentation
- Attackers probe IPv6 when IPv4 is locked down β€” don't neglect it

## Debugging
- Test from outside your network β€” rules may look correct but not work
- Provider dashboards often show blocked traffic logs
- "Connection refused" = port closed properly; "Connection timeout" = firewall dropping silently
- Online port scanners verify what's actually open from the internet

## Common Mistakes
- Opening ports "temporarily" and forgetting to close them
- Opening 80/443 when no web server runs β€” unnecessary exposure
- Forgetting UDP for services that need it β€” DNS, VPN, game servers
- Assuming firewall is active β€” verify it's actually running/applied
- Only configuring IPv4 β€” leaving IPv6 wide open
- Trusting "security through obscurity" β€” non-standard ports slow attackers, don't stop them

Discuss this resource

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.