MCP Email Verification for AI Agents

By Aria Pramesi, founder of InboxPolicy · Updated July 4, 2026

InboxPolicy ships an MCP server exposing decide_send, verify_email, batch, and usage as agent tools. An agent asks should I email this address and gets back one of five actions (send, send_with_caution, review, retry_later, avoid), a confidence score, normalized signals, and the cost, in a single tool call, no dashboard required; hosted MCP calls require an API key.

What tools does the InboxPolicy MCP server expose?

Seven tools, all callable directly by an agent without a REST client or dashboard:

• decide_send: returns one of five policy actions plus confidence, status, normalized signals, cache state, and cost.
• verify_email: returns detailed verification fields, including status, deliverability, reachability, MX context, normalized signals, confidence, provider/cache metadata, and the policy-derived recommended_action.
• create_verify_batch, get_batch_status, and get_batch_items: create and inspect asynchronous batches with per-item results.
• estimate_credits: estimates the credits required for a batch.
• get_usage: reports usage and remaining credits.

How do I connect the MCP server to my agent?

InboxPolicy runs a hosted MCP server over Streamable HTTP — no local install. Add it to your agent's mcpServers config block with your API key as a header:

{
  "mcpServers": {
    "inboxpolicy": {
      "url": "https://mcp.inboxpolicy.com",
      "headers": { "x-api-key": "ip_live_..." }
    }
  }
}

Once connected, the agent sees decide_send, verify_email, create_verify_batch, get_batch_status, get_batch_items, estimate_credits, and get_usage as first-class tools it can call mid-conversation. Get a key by buying a credit pack at inboxpolicy.com/#pricing.

What does an agent get back from a single call?

A decide_send call returns the decision fields the agent needs, including normalized signal booleans and cost, for example:

{
  "request_id": "vrf_...",
  "email": "[email protected]",
  "action": "review",
  "reason": "Catch-all mailbox detected. Review before sending.",
  "confidence": 60,
  "status": "catch_all",
  "retry_at": null,
  "signals": {
    "catch_all": true,
    "disposable": false,
    "free": false,
    "role_account": false
  },
  "from_cache": false,
  "cost": { "credits": 1, "billable": true }
}

The agent does not have to parse a status enum and guess what it means for send timing. It gets the decision directly.

How does pricing work for agent calls?

The hosted MCP server requires an x-api-key backed by prepaid credits. Keyless $0.01 pay-per-call is available separately through the REST endpoints over x402: a REST request gets HTTP 402 payment requirements, the client attaches a PAYMENT-SIGNATURE header, and receives the decision plus an on-chain settlement receipt.

For agents or teams that prefer paying by card up front, prepaid credit packs are available: Starter $5 for 1,000 credits ($5.00/1k), Builder $19 for 5,000 ($3.80/1k), Growth $79 for 25,000 ($3.16/1k). Volume pricing is available on request. Two result types can cost 0 credits on eligible calls: a cache hit for the same normalized address and strictness within 72 hours when force_refresh is false, and malformed-email rejection before SMTP. For API-key calls, matching idempotency replays are limited to the record TTL; keyless x402 does not use that wrapper.

There is no free tier. The $0.01 x402 call is the trial, on purpose, since free tiers tend to attract list-cleaning abuse rather than genuine agent use.

How accurate are the decisions?

On InboxPolicy's prior verification engine, valid-verdict agreement with MillionVerifier was roughly 90% typical across more than 2 million verifications, though this is not a formal recent benchmark and it varies by vertical, dropping as low as 60% in some.

Unknown results map to review. Catch-all results map to review by default or send_with_caution under aggressive strictness. This matters for agents because roughly 30-40% of B2B addresses sit on catch-all domains, where SMTP alone cannot confirm a mailbox exists.

When should an agent use a different tool instead?

InboxPolicy is built for a live send decision inside an agent workflow, not every verification job:

• One-shot cleaning of a huge scraped list at the lowest per-address cost at high volume: MillionVerifier's $3.90/1k entry (down to $0.45/1k at 1M volume) pricing is built for exactly that. InboxPolicy's 25,000-credit Growth pack has a lower unit price at $3.16/1k, while MillionVerifier's 10,000-credit entry purchase costs less overall at $39.
• Spam-trap and abuse-address detection: ZeroBounce maintains a dedicated spam-trap and abuse database that InboxPolicy does not.
• A dashboard-first marketing suite with CSV uploads and integrations: Kickbox or Emailable are built around that workflow rather than an agent tool call.

Frequently asked questions