Hi everyone. I’m trying to set up my first OpenClaw agent with a few plugins, like the web search and file writer. My goal is to lock down its network access for privacy/security, but every time I configure egress rules in my firewall, something stops working.
I’m nervous about breaking things. Could someone explain what domains or IP ranges the common plugins actually need to function? I’m especially unsure about the AI service APIs (I’m using OpenAI and Anthropic) and any external data fetches.
Is there a baseline “allow list” I should start with, or a way to test what connections are being attempted? Any guidance would be so appreciated. 😅
You're right to be cautious. A generic allow list is problematic because plugin dependencies can change. The correct approach is to monitor egress traffic first.
Run your agent with firewall logging enabled for a blocked default policy, or use a tool like tcpdump on the agent's host during a typical workflow. That will show you the exact FQDNs the plugins attempt to reach for your specific configuration. For the AI APIs, you'll see calls to domains like api.openai.com and api.anthropic.com, but also potentially their CDNs for status or auxiliary services.
Start by allowing those observed domains, then incrementally test each plugin's function. The web search plugin is particularly brittle, as it may follow redirects to arbitrary domains. You'll need to decide if you'll allow outbound port 80/443 broadly for that plugin, or accept limited functionality. NIST SP 800-41 revision 1 has useful guidance on application-layer filtering that applies here.
Policy is code