Outbound connections and DNS

The short version:

Your instance can reach the internet, and name resolution is provided by the platform. You do not configure it, and there is nothing to switch on.

If an HTTP Request node, an outbound webhook or a credential test cannot reach a web address, this page is the place to start.

What your instance can reach

Outbound HTTPS and HTTP	Any public address, on the normal ports
Web addresses by name	api.example.com resolves the same as it would anywhere
Third-party credential tests	Slack, Gmail, Airtable, Notion and the rest connect normally
Inbound to your instance	Your own HTTPS address, which is a separate thing - see What we manage

You do not need to allowlist anything, open a port, or tell us which services you plan to call.

Where DNS lookups go

Name lookups from your instance are answered by Cloudflare (1.1.1.1), with Google (8.8.8.8) as a fallback.

We are telling you this plainly because it is worth knowing: the name you look up - not the contents of the request - is seen by a third-party resolver. If your organisation has a policy about that, or a security questionnaire asks, this is the answer. The connection itself goes directly from your instance to the service you are calling.

This is a platform setting. It is not configurable per instance, and you cannot point your instance at your own DNS server.

If a request is failing

The exact error text matters, because two very different faults look similar.

Errors that mean a name could not be resolved:

• EAI_AGAIN
• getaddrinfo ENOTFOUND
• "The service refused the connection"
• "Couldn't connect with these settings" on a credential test
• "The HTTP Request node fails but the URL works in my browser"

Check whether every address fails or only one:

• One address fails, others work. The problem is at the other end - that host is down, blocking us, or the address has a typo. Try the same URL from your own machine.
• Every address fails. That is ours. Open a ticket and include the error text exactly as it appears.

If the workflow instead runs for a while and then stops, that is a different thing entirely - it is the run limit on your plan, not a connection problem. See Choosing a plan.