@rail402.dev/safe-fetch

Outbound HTTP for URLs that come from untrusted input, such as a resource URL an AI agent picked or a seller's origin.

createSafeFetch() returns a fetch-compatible function built on node:http(s), with no third-party dependencies (only @rail402.dev/errors, which has none). It returns a native Response, or rejects with an EgressError that carries a stable code, a stage and a non-empty reason:

• https only on port 443 by default; http (port 80) only when enabled.
• The URL is checked before any DNS lookup. IP literals are classified here, in every spelling, because Node does not call a custom lookup for them.
• Private, loopback, link-local, carrier-grade NAT, cloud metadata, documentation, NAT64, 6to4 and IPv4-mapped addresses are refused. allowPrivateNetwork: "loopback" admits this machine only, for local development.
• DNS answers are checked when connecting and the socket dials only validated addresses; a host with any private address is refused whole. The connected address is checked again. Environment proxies are never used.
• Redirects are refused unless a request opts into redirect: "same-origin", which follows up to 3 hops on the same host and port.
• Request headers and bodies, response headers and decoded response bodies (gzip, deflate, br) are capped. Connect, header, idle and total timeouts apply.
• Errors never repeat a resolved address.

import { createSafeFetch } from "@rail402.dev/safe-fetch";

const safeFetch = createSafeFetch();
const response = await safeFetch("https://api.example.com/weather?city=Paris");

npm install @rail402.dev/safe-fetch

Part of Rail402, x402 payments and discovery for Stellar. Documentation: docs.rail402.dev. Requires Node.js 24.11 or later. Apache-2.0.