How to use MCP to validate emails

Validate email addresses from Claude, Cursor or VS Code with isBusinessEmail's MCP server: classify with check_email, confirm mailboxes with verify_email.

  • #mcp
  • #verification
  • #api
A chat message asks to verify jane@acme.io; the assistant calls the verify_email tool and gets status deliverable, reason mailbox_exists, then answers that it is safe to send. Verify that [email protected] exists tool call · isbusinessemail verify_email { "email": "[email protected]" } status deliverable reason mailbox_exists is_catch_all false confidence 0.97 Yes: Acme's mail server accepts [email protected] and rejects made-up addresses. safe to send

You can validate email addresses from an AI assistant by connecting it to an MCP server that does the checking. The model can’t resolve DNS or talk to mail servers itself, but it can call tools: once isBusinessEmail’s MCP server is connected, “is this email valid?” in Claude, ChatGPT, Cursor or VS Code runs a real check and comes back with the answer and the reasons.

This post is about validation: catching bad, fake and undeliverable addresses. If you’re mostly interested in whether an address is a work email and which suite the company runs, checking business emails from your AI assistant covers that side.

Two kinds of email validation

“Is this email valid?” can mean two things, and the server has a tool for each:

Question Tool How it answers Speed
Is this a real, usable address? Typo, dead domain, no mail server, disposable inbox? check_email, check_domain, check_batch Syntax, typo suggestions, DNS and MX records, curated lists. Never contacts a mail server Instant
Does this exact mailbox exist? verify_email Asks the domain’s mail server over SMTP, plus a catch-all test. No email is sent A few seconds

Use the first kind for everything, and the second for the addresses you’re about to email. Email verification vs classification explains why both exist.

Connect your assistant

The server is at https://api.isbusinessemail.com/mcp. You sign in with your isBusinessEmail account (a free one is fine), and each connected app gets its own API key that you can revoke.

Claude (web, desktop and mobile): open Settings → Connectors, choose Add custom connector, name it isBusinessEmail, paste the URL, then Connect, sign in and Allow access.

Claude Code:

claude mcp add --transport http isbusinessemail https://api.isbusinessemail.com/mcp

Then run /mcp and choose Authenticate.

Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "isbusinessemail": { "url": "https://api.isbusinessemail.com/mcp" }
  }
}

VS Code (.vscode/mcp.json):

{
  "servers": {
    "isbusinessemail": { "type": "http", "url": "https://api.isbusinessemail.com/mcp" }
  }
}

ChatGPT and clients that can’t sign in with OAuth are covered in the MCP docs, including the API key option. To confirm the connection, ask “How many checks do I have left today?”: get_usage is free.

Validate addresses: check_email and check_batch

Start with classification. It catches most bad addresses without contacting anyone’s mail server:

Is [email protected] a valid email?

The assistant calls check_email and gets back the category, a did_you_mean suggestion (gmail.com here), a recommendation and the reasons. For a domain that doesn’t exist or has no mail server, category is invalid; for a throwaway inbox it’s disposable.

For a list, paste it and ask for a table:

Here are 60 email addresses from our event sign-up sheet. Use check_batch and give me a table with email, category, did_you_mean and recommendation. Then list the invalid and disposable ones separately.

check_batch takes up to 100 addresses per call and returns them in input order, which keeps the table lined up with your sheet.

Confirm the mailbox exists: verify_email

Classification can’t tell you whether [email protected] is a real person or a typo for [email protected]: both are on the same company domain. For that, the assistant needs verify_email:

Before I send this proposal, verify that [email protected] actually exists.

The tool connects to Acme’s mail server, asks whether it accepts [email protected], asks about a random address to rule out a catch-all domain, and hangs up before anything is sent. The assistant gets a short summary such as:

[email protected]: deliverable (mailbox_exists) — the mail server accepts this address and rejects made-up ones at the same domain. Confidence 97%.
Category: business. Mail provider: google_workspace.

and the full result as structured content:

{
  "status": "deliverable",
  "reason": "mailbox_exists",
  "confidence": 0.97,
  "mailbox": "exists",
  "is_catch_all": false,
  "category": "business",
  "did_you_mean": null
}

The four statuses mean:

Status Meaning What to do
deliverable The mailbox exists and accepts mail Send
undeliverable No such mailbox, a disabled one, or a domain that can’t receive mail Don’t send; look for a typo
risky Catch-all domain, full mailbox or disposable inbox Send with care, or confirm another way
unknown The mail server greylisted the check, timed out or refused to answer Ask again later; never treat it as invalid

When the reason is greylisted, the server asked us to come back later. We retry in the background, so asking the assistant again a few minutes later usually returns the real answer. The bounce codes reference explains how server replies map to these reasons, and the verify endpoint docs list every reason code.

Prompts that work well

  • “Check these 40 addresses with check_batch. Then run verify_email only on the business addresses we plan to email, and show status and reason for each.”
  • “Verify [email protected]. If it’s undeliverable, check whether there’s a likely typo.”
  • “Which of these addresses are on catch-all domains?”
  • “Is this a disposable email? [email protected]”
  • “How many checks do I have left today?”

Two habits keep the answers trustworthy:

  1. Ask for the fields. “Show status, reason and confidence” makes the assistant quote the result instead of paraphrasing it.
  2. Classify first, verify second. There’s no point verifying a disposable inbox you’re going to drop, and each verification counts against a daily cap.

Limits

  • Every tool call counts toward your account’s quota like an API call: check_email, check_domain and verify_email count 1 check, check_batch 1 per item, and get_usage is free.
  • verify_email also has its own daily cap, because each one is an SMTP conversation: 100 a day on the free tier, then 500, 2,000 and 10,000 on paid tiers.
  • verify_email takes a few seconds per address, sometimes up to about 30 for a slow mail server. For thousands of addresses, call the API from a script instead of a chat.

You can try every status without spending checks on test addresses such as [email protected] or [email protected]; the docs list them all.

Privacy

The MCP server follows the same rules as the API. Classification tools never contact a mail server. verify_email asks the domain’s mail server about that one address and sends no message, so the person isn’t contacted. We don’t store the address: results are cached under a keyed hash, and the server’s reply text is returned with addresses and IP addresses removed. Nothing is ever asked of Google or Microsoft about a person. Privacy and data has the details.

Your AI provider is also in the loop: whatever you paste goes to the assistant before any tool runs. Only paste addresses you’re allowed to share with it, remove columns you don’t need, and keep tool approval switched on in your client.

Get started

  1. Create a free account, or let the sign-in screen create one when you connect.
  2. Add https://api.isbusinessemail.com/mcp to your assistant using the steps above or on the MCP page.
  3. Ask it to verify [email protected]. It’s free, and it shows the full response.
  4. Then try a real address you’re about to email.

Prefer a form to a chat? The free email verifier runs the same check in your browser.

Frequently asked questions

Can an AI assistant validate email addresses?

Yes, if it can call a tool that does the checking. Language models can't look up DNS or talk to mail servers on their own, but through an MCP server they can call a validation service and report its answer.

What is the difference between check_email and verify_email?

check_email classifies the address from its domain, DNS and curated lists: is it valid, disposable, business or personal. It never contacts a mail server and answers instantly. verify_email asks the domain's mail server whether this exact mailbox exists, which takes a few seconds and has its own daily cap.

Does verify_email send an email to the address?

No. It opens an SMTP conversation with the mail server, asks about the address and a random address at the same domain, and disconnects before any message is sent.

Can I validate a whole list with MCP?

check_batch classifies up to 100 addresses or domains per call. verify_email checks one address per call, so for a list, ask the assistant to classify everything first and verify only the addresses you plan to email.

Which AI assistants support the isBusinessEmail MCP server?

Any client that can connect to a remote MCP server over Streamable HTTP. The docs have setup steps for Claude, Claude Code, ChatGPT, Cursor and VS Code, plus an API key option for clients without OAuth.