AI SDRs are genuinely useful: they research accounts, personalize outreach, and follow up tirelessly. But they have a failure mode human SDRs don't — they make mistakes at machine speed. An agent that guesses an email pattern, enriches a stale record, or hallucinates a contact will happily send hundreds of messages to dead inboxes overnight. You wake up to a bounce spike, a spam-placement problem, and a domain that took months to warm up, now burned.
A human SDR might send 50–100 emails a day and notice when replies dry up. An AI agent sends 10x that without noticing anything. At a 5% bounce rate on 1,000 sends, that's 50 hard bounces in a day — enough to get inbox providers' attention for all the wrong reasons. Verification before every send turns that 5% into ~0%.
The fix isn't to slow the agent down — it's to give it a guardrail: no draft, no send, until the recipient verifies. This works best as a native tool call, not a separate process the agent might skip. That's exactly what our MCP integration is for.
Connect verify_email as a native tool in Claude Code, Cursor, or any MCP-capable client — it's a config snippet, not a project. Then prompt the agent: “call verify_email for every recipient before drafting; skip invalid addresses and flag risky ones for review.” The tool returns a verdict, reasons, per-layer checks, and remaining credits. Full setup walkthrough: Guide 1.
Building your own agent loop? POST /api/verify returns a job; poll until it's done and act on the verdict — valid sends, invalid drops, risky/unknown goes to review. Copy-paste Python and JavaScript examples handle polling, 402 (out of credits), and 429 (rate limit) for you. The API also runs the full live SMTP mailbox handshake, which the MCP tool skips.
If your agent works from AI-generated lead lists — Clay, Apollo, or your own prospector — bulk-clean the CSV before it ever reaches the agent. Upload up to 10,000 rows, download verdicts per row, and hand the agent a list that's already safe. A 5,000-lead clean costs a $30 one-time pack, and unused credits never expire.
One credit per verification — the same pricing as everything else: 100 free every month, plans from $15/month, packs from $7. Compared to the cost of a burned domain (new domain, new warmup, weeks of lost pipeline), it's a rounding error.
Treat unverified AI outreach like unreviewed AI code: never ship it. Wire verify_email into the workflow once, and every send after that is guarded — while you sleep, at machine speed, without thinking about it. Set it up in minutes.
100 verifications every month. Five-layer pipeline, API, bulk CSV, and MCP for AI agents.
Start freeNo credit card required.
Everything else, ask us.
Afterward is too late — the bounces already happened and already hurt your reputation. Verification is preventive; bounce handling is damage control. Do the former so you rarely need the latter.
The MCP tool runs syntax, disposable-domain, MX, and role-account checks — verdicts top out at “unknown”/“risky”. For the definitive live SMTP handshake on high-value sends, escalate to the REST API.
One credit per email — fractions of a cent on paid plans, and the first 100 each month are free. It's the cheapest insurance in your outbound stack.