Dedicated egress IP

What it is

Egress is outbound traffic. Cloud platforms usually send outbound requests from a pool of shared addresses that change over time. A dedicated egress IP is one fixed address, used only by you, that every outbound request leaves from.

Why it matters

Many systems only accept connections from known addresses: a vendor API with an IP allow-list, a customer's firewall, a managed database. If an agent platform calls them from shared, changing addresses, you either open the allow-list wide, which defeats it, or the calls fail. A dedicated address lets you allow exactly one IP and know that only your traffic uses it.

How it works

1. The platform reserves a static public address for the tenant.
2. Outbound requests for that tenant are routed through a hop that owns the address, often using an HTTP CONNECT tunnel (RFC 9110 section 9.3.6) so TLS runs end to end between the platform and the vendor.
3. The vendor sees every request from that one address.
4. If the hop is unavailable, a fail-closed design refuses the request rather than sending it from a shared address.

The MCP security guidance recommends egress proxies for server-side clients more generally, as a place to enforce network policy.

Example

A finance team's database only accepts connections from listed addresses. They add the tenant's dedicated IP to the database firewall. The agent's read-only queries now reach the database, and nothing else on the platform can.

Common pitfalls

• "Static" but shared. An address shared with other customers lets their traffic through your allow-list too.
• Failing open. Falling back to a shared address when the dedicated path is down silently breaks the guarantee.
• HTTPS only. Databases speak their own protocols; check they are carried.
• Losing the address on re-provisioning, which forces every allow-list to be updated.

How actionlake handles it

The actionlake Dedicated IP add-on gives a tenant one static IPv4 address that is not shared with any other tenant, in the Sydney region. It fails closed: if the address cannot be used, requests are refused, never sent from a shared one. It carries HTTPS and MySQL, PostgreSQL and Redis, with the database's own TLS end to end, on ports 443, 80 and 1024 to 65535. Not yet carried: plain HTTP and other ports below 1024. It is available on the Team plan and above and added on request, and Enterprise includes one; when the add-on is removed, the address is held for 30 days. The price is on the pricing page; ask through contact.