Let your AI agent test email flows end to end

July 30, 2026

An AI agent building or testing your app hits a wall at exactly one place: the verification email. It can fill the signup form, it can click the button, and then the flow leaves the browser. The code it needs is sitting in an inbox it does not have.

The usual workarounds are all bad in a different way:

• Mock the mailer and you are no longer testing what you shipped. The template bug, the broken link, the SPF failure all slide through.
• A shared team inbox means credentials in agent context and tests that trample each other’s messages.
• Copy the code over yourself and it is not automation anymore. You are the transport layer.

The missing tool is an inbox

InboxDome runs a hosted MCP server at https://inboxdome.com/mcp. Connect it and the inbox becomes a set of native tools in the agent’s hands:

1. create_inbox — a fresh, real address for this test run
2. wait_for_email with waitFor: "otp" — blocks until the mail lands, returns the extracted code
3. run_assertions — delivery, content, links, SPF/DKIM/DMARC, pass or fail

So “test the signup flow” becomes something an agent completes alone: it mints an address, registers with it, waits for the OTP, submits it, then asserts the welcome email actually arrived and authenticates. Real SMTP the whole way; nothing mocked.

Connect it

Claude Code:

claude mcp add --transport http inboxdome https://inboxdome.com/mcp \
  --header "Authorization: Bearer idk_live_..."

Any MCP client:

{
  "mcpServers": {
    "inboxdome": {
      "type": "http",
      "url": "https://inboxdome.com/mcp",
      "headers": { "Authorization": "Bearer idk_live_..." }
    }
  }
}

No key yet? Connecting without one still initializes and lists the tools, and the first tool call tells the agent how to register itself: one POST /v1/accounts with an email returns a free key, 100 inbox tests a month, no card. An agent can do the whole setup without you leaving the conversation.

Why receive-only matters here

An InboxDome address can never send anything. For agent use that is a feature twice over: a misbehaving agent cannot spam anyone from your test infrastructure, and there is nothing in the inbox worth stealing after the run. Inboxes delete themselves.

The same boundary applies to purpose: this is for testing applications you own or are authorized to test — not for creating accounts on other people’s services. The Acceptable Use Policy is short and enforced.

Full tool reference in the MCP guide; the same capabilities are in the HTTP API if you’d rather script it yourself.

Get a free API key: 100 inbox tests a month, no card. Orgrab a temporary inbox without signing up at all.