Tool Thursday: MillionVerifier, the Email Check My Agent Runs Before Every Send (26 of My First 54 Outreach Emails Bounced)

Every Thursday I pull one tool out of my AI agent stack and show you how it actually gets used. This week it’s the MillionVerifier API. It answers one question: will an email address accept mail? My outreach agent asks it before every single send. I learned that habit the expensive way.

The receipt: 26 of my first 54 outreach emails bounced

My agents run a backlink outreach campaign. One agent finds blogs and pulls contact addresses off their pages. Another loads 5 of those addresses a day into Instantly, which sends the pitch.

In July, the first 54 addresses went in between July 10 and July 18. 26 of them bounced. That’s 48%. Every one of the 26 was marked “Verified” in my lead table. When I went back to check, 19 of them had no verification result saved at all. The label was there, but no evidence came with it.

Since July 18, the campaign has loaded 467 more addresses and sent 320 more emails. Bounces since then: 0. Lifetime, that’s 26 out of 413 sends, and all 26 came from that first week.

What changed: check right before sending

Two habits. I can’t prove which one did more of the work, but together they took bounces to zero:

1. The sending agent re-checks every address minutes before it goes out. It doesn’t trust a check from days ago. Addresses go stale fast. On September 1, one address passed at 1:58 PM and came back “unknown” at 5:58 PM the same day. The agent dropped it and sent to the next one instead.
2. Every verdict gets saved with its raw result. Now “Verified” means “MillionVerifier said ok, and here’s the result,” not just a word somebody typed into a column.

The finder agent also does its own check when it saves an address. Across 1,363 addresses found on real websites, it marked 175 invalid (13%) and 147 risky. None of those got a pitch.

How the MillionVerifier API works

It’s one GET request per address:

curl -s -A "Mozilla/5.0" \
  "https://api.millionverifier.com/api/v3/?api=YOUR_KEY&email=hello@example.com&timeout=10"

Here are the three things my agent had to learn the hard way:

• Route on result, not quality. The response has both. result is ok, catch_all, unknown, invalid, or disposable. A catch-all shows up as quality: risky, so code that branches on quality threw away 3 of 10 good leads in one run before the agent caught it.
• “Unknown” isn’t a no. If the subresult is timeout, greylisted, or ip_blocked, that’s a problem on the checker’s side. It tells you nothing about the mailbox. My agent retries once with a 20-second timeout before it decides anything.
• Send a browser User-Agent. A plain Python request gets HTTP 403 before a single address is checked. The same URL works fine from curl.

Catch-all domains accept mail for any address, so MillionVerifier can’t confirm the mailbox exists. My agent sends those to a second checker, BounceBan, for a second opinion. That’s the “waterfall”: the cheap check first, and the second checker only for the addresses the first one couldn’t call.

What it costs

You get 100 free credits when you sign up. After that you buy credits upfront: $39 for 10,000, or $89 for 50,000 (prices from their site on October 29). The credits never expire, and one credit checks one address. My account has 39,908 left. A batch of 5 outreach emails a day barely touches that.

Compare that to the cost of a bounce. Every hard bounce tells inbox providers your sending address might be a spammer. A week of 48% bounces can hurt deliverability for every email you send after it, including the ones to real customers.

The takeaway

If an agent sends email for you, give it one rule: check the address right before you send, and save the answer. A “Verified” label with no result attached is a guess. Want to see the rest of the stack? It’s all mapped out in what my agentic OS looks like after 139 days.

If your agents send email and you’re not sure what they’re checking first, book a strategy session and we’ll go through it together.

See you next Thursday.
Jon