mailifier

Ask a mail server whether an address exists. No mail is ever sent.

This is the handshake the paid verification services run, from a host of yours: MX lookup → connect on port 25 → EHLO → STARTTLS when offered → MAIL FROM → RCPT TO <the address> → QUIT. The server's answer to RCPT is the verdict. There is no DATA, so nothing is ever delivered.

One runtime dependency: tldts, for the public suffix list. Node 22+.

Install

pnpm add mailifier

Use it

import { LocalChecker, SmtpProbe } from "mailifier";

// Stage 1: free, no sockets. Syntax, typosquats, disposable providers, MX routing.
const local = await new LocalChecker().check("[email protected]");
if (!local.passed) return local.failure; // "no_mx", "syntax: …", "disposable_domain"

// Stage 2: ask the mail server.
const probe = new SmtpProbe({ helo: "probe.example.com" });
const verdict = await probe.verify("[email protected]");
// { result: "valid", raw: { reason: "accepted", mx: "…", code: 250, transcript: [ … ] } }

The four answers

| result | means | | --- | --- | | valid | the server accepted the address | | invalid | the server refused it by name, or the domain takes no mail at all | | catch_all | the domain accepts everything, so acceptance proves nothing | | risky | nobody would tell us: greylisting, a refused connection, a policy block |

Only valid and invalid are facts. catch_all is common on Microsoft 365 tenants. risky is about the conversation, never about the address — retry it later.

Running it somewhere that can

Most clouds block outbound port 25 until you ask them to stop. So the usual shape is one small host that can talk SMTP, and everything else asking it over HTTP.

On that host:

PROBE_TOKEN=… PROBE_HELO=probe.example.com npx mailifier

Everywhere else:

import { RemoteProbe } from "mailifier";
const probe = new RemoteProbe(process.env.PROBE_URL, process.env.PROBE_TOKEN);

Both satisfy MailboxProbe, so the calling code never knows which one it has.

The server is bearer-authenticated, caps calls in flight (PROBE_MAX_IN_FLIGHT, default 8; beyond it 429; RemoteProbe waits and retries a 429 for about a minute before it gives up), and runs a canary before every verdict: if port 25 is not open from this host it answers 503, never a risky verdict. A closed port is the host's problem, not the mailbox's — and a verifier that quietly reports risky for a working mailbox is worse than one that stops.

Each MX host gets one conversation at a time with a gap between probes (PROBE_HOST_GAP_MS, default 1500). Google, Microsoft, Proofpoint and Mimecast front thousands of domains under one host name, so a list heavy in them queues behind that one host; PROBE_BIG_HOST_LANES (default 1) gives those fleets that many conversations at once, each with its own gap.

GET /healthz → { ok, in_flight, port_25 }.

The published package also carries dist/mailifier.mjs: the whole server in one file, to drop on a box that has node and no npm. pnpm bundle rebuilds it.

Being a good guest

• One conversation per MX at a time, with a gap between them (perHostGapMs, 1.5s).
• helo should be a name whose forward and reverse DNS agree, and MAIL FROM is postmaster@ that name, so a curious postmaster can look you up. Use a name you do not send real mail from.
• At most three MX hosts tried per address.
• DNS goes over HTTPS (Cloudflare), so it works in a container with no resolver.

Interfaces

MailboxProbe is the only thing worth depending on:

interface MailboxProbe {
  readonly name: string;
  verify(email: string): Promise<Verdict>;
}

Deliberately not on it: whether a verdict may be trusted, and whether it costs money. That is your policy about a probe, not something a probe knows about itself.

Every network edge is injectable — dial, resolver, fetch, sleep — so the tests run with no sockets and no DNS.

Testing against it

src/smtp.test.ts shows the pattern: hand probeMailbox a scripted Conversation and assert the verdict it reads out of the replies.