Email checks for your agents.

Give an agent a specific verification tool, a bounded credit budget and structured evidence. Keep sending decisions in your application.

Tasks with a clear stopping point

Give your agent evidence it can act on.

Contact intake

A customer supplies an address during a chat. The agent checks it, explains a formatting error and asks for a correction. It keeps the original input until the customer confirms a replacement.

Import review

An agent reviews a supplied set of contacts within a fixed check budget. It groups invalid, risky and unknown results separately and returns a review report for a person to approve.

Onboarding triage

A business cannot receive account mail. The agent checks the supplied address and reports missing mail routing or an inconclusive probe, with a next action for the account owner.

One tool call. Evidence first.

  1. Read a user-supplied address.
  2. Call verify_email within the budget.
  3. Inspect status and null checks.
  4. Return a diagnosis or request review.

The connector exposes one tool: verify_email. It does not search for addresses, edit a CRM or send mail. Your host controls any later action.

Two ways to integrate

Use the OpenAPI definition to generate a client or tool schema, or run the included MCP connector from the connector download. Both call the same authenticated API with the same credit accounting.

Download and unzip emailvalidly-mcp-1.0.0.zip
cd emailvalidly-mcp
npm ci
# Set EMAILVALIDLY_API_KEY using your secret manager.
# Optional: EMAILVALIDLY_MAX_CHECKS=25
npm start

The connector uses stdio transport. Configure your MCP host to launch Node with the absolute path to mcp.js and provide EMAILVALIDLY_API_KEY through its environment or secret settings. It requires a Growth or Pro API key once paid plans launch. It is not a public remote MCP endpoint.

A small, explicit tool contract

{
  "name": "verify_email",
  "arguments": {
    "email": "person@example.com",
    "deep": false,
    "request_id": "crm-contact-unique-event-001"
  }
}

Reuse request_id when retrying the same operation. A fresh verification needs a fresh identifier. The connector defaults to a maximum of 100 unique checks per process and accepts an EMAILVALIDLY_MAX_CHECKS setting from 1 to 1,000. This is a local session guard; your account balance and server rate limits still apply.

Useful agent workflows

Verification is a diagnostic step. The MCP connector has no send-email capability. An agent should not infer consent from a valid address, invent an inbox owner, or automatically discard an unknown result.

Failure handling that keeps costs predictable

401 means the key or plan needs attention; 402 means credits are exhausted. Respect Retry-After on 429. Retry a network failure or an in-flight 409 with the same request_id. Never retry by generating a new identifier unless you intend to spend another credit.

Read the full API reference →

Configure a local MCP host

After installing the downloaded connector, adapt this configuration to your host. Replace the path with the absolute location of mcp.js. Supply the token through the host’s secret settings; the placeholder below is not a working credential.

{
  "mcpServers": {
    "emailvalidly": {
      "command": "node",
      "args": ["/absolute/path/emailvalidly-mcp/mcp.js"],
      "env": {
        "EMAILVALIDLY_API_KEY": "YOUR_SERVER_SIDE_TOKEN",
        "EMAILVALIDLY_MAX_CHECKS": "25"
      }
    }
  }
}

A task instruction you can adapt

Check only the addresses supplied in this task.
Use at most 25 unique verification requests.
Reuse each request_id if a request needs a retry.
Report invalid, risky and unknown results separately.
Keep inconclusive checks visible. Do not send any email.
Stop when the budget is reached and report unprocessed rows.

Count unique operations in your host as well as the connector. Restarting the connector resets its local counter; it does not reset account credits. Choose request identifiers of 8–100 letters, digits, underscores or hyphens.

Agent integration questions

Can I connect over a hosted MCP URL?

No. This release runs locally over stdio. For a hosted agent, run the connector alongside your worker or call the HTTP API from your server.

Does a valid result authorise sending?

No. Mailbox evidence says nothing about consent. Keep any send action separate and subject to your application’s permissions.

Can I use it before paid plans launch?

You can download and inspect the connector now. Authenticated verification needs an active Growth or Pro plan, which is not yet open for purchase. The public free checker is available today.

Try a real address first.

See the checks in the free tool, then use the same evidence in your workflow.