What is an AI agent email address?

It is a working email inbox that an AI agent — a coding assistant, an automation or an LLM tool — can read and use on its own, usually through an API or the MCP protocol, instead of a human opening a browser tab.

How does an AI agent read its email?

Tempx exposes a scoped MCP server: the agent calls list, read, search and wait_for_email tools. Pairing happens in your browser with one approval; each connection is limited to a single inbox and can be revoked in one click.

Can an AI agent send email?

Agents can submit mail over Tempx's capture-only SMTP gateway, where every message is stored in the paired inbox for inspection instead of being delivered — a safe way to test outgoing email without contacting real people.

Which agents work with Tempx email?

Any MCP client: Claude Code and Claude Desktop pair through the browser, ChatGPT connectors work the same way, Z.ai and custom automations authenticate with a Bearer token minted from the account page.

Email for AI Agents

An AI agent email address is a working inbox that belongs to your agent, not to you: the agent reads it over an API, waits for the messages it needs, and uses it to complete real workflows. Tempx gives every agent such an inbox — with a scoped MCP email server, a capture-only SMTP testing gateway and no human sign-up in the loop.

If you build with LLMs, you already know the pattern: the agent gets impressively far, then hits “check your inbox to verify”. Email for AI closes that gap. Connect Claude, ChatGPT or Z.ai in one step, or read on for what agent email should look like.

Why agents need their own email

Verification codes. Signups, logins and one-time passwords arrive by mail. An agent with its own inbox receives the code directly in the loop — wait_for_email hands it over the moment it lands.
End-to-end testing. To test a signup flow, the agent needs both sides: submit the form, then read the confirmation. An agent email address makes the whole cycle autonomous.
Reading notifications. Webhooks fail, feeds need polling — but nearly every service still sends email. For an agent, a mailbox is the most reliable integration surface on the internet.
Isolation. The agent's mail never touches your personal inbox. When the task is done, the address (and everything in it) simply expires.

The MCP email server

Tempx ships a scoped MCP server over HTTPS — an email API designed for agents rather than for humans — so any MCP client becomes an email-capable agent. Tools: list recent messages, read a message, search the inbox, and wait_for_email — a blocking wait that returns as soon as a matching message arrives, which is exactly what verification-code flows need.

Pairing is browser-based: when the agent connects, your browser opens tempx.uk, you approve one inbox, one permission set and one expiry. No token is ever pasted into a chat or a shell. Every connection can be revoked from the account page in a single click, and only a hash of the credential is stored.

claude mcp add --transport http tempx https://tempx.uk/mcp

Email testing: sending that never leaves the lab

Agent workflows that send mail are handled by the capture-only SMTP gateway at smtp.tempx.uk:587 (STARTTLS). The agent or your application authenticates with credentials minted from the account page, “sends” like it would with any mail server — and every message is captured into the paired Tempx inbox for inspection, in the Mailtrap tradition. Nothing is ever relayed to a real recipient, so an agent test run can never email a real person.

In short: receiving is open to any sender anywhere; sending is a capture-only sandbox whose output lands in exactly the inbox your credentials point at. Read side and send side share one inbox, which makes full-loop tests simple: the agent submits a signup form, receives the verification mail, and verifies the outgoing notification — all against disposable addresses.

What AI agent email looks like on Tempx

	Tempx agent email	Your personal inbox
Access	MCP tools + API, made for agents	Webmail made for humans
Waiting for mail	`wait_for_email` blocking tool	You refresh the tab
Sending	Captured for testing, never delivered	Really sends
Scope	One inbox per token, revocable	Your entire mailbox
Lifetime	Minutes to days, auto-deleted	Years

Who uses agent email

Coding agents (Claude Code, Claude Desktop) that test signup and password-reset flows end-to-end.
ChatGPT and Z.ai users who want the assistant to receive its own verification codes instead of dictating them.
QA and CI pipelines that need a fresh mailbox per run — agents can mint SMTP credentials themselves.
Automations that watch a mailbox instead of building a webhook for every service.

Get started

Open the agent guide for Claude, ChatGPT or Z.ai — one command or one connector.
Approve the inbox in your browser; the agent starts with list and wait_for_email.
Need sending? Mint SMTP credentials on the account page and point your mailer at the capture gateway.

Give your agent an inbox

One approval in the browser and your AI agent has its own working email address.

Connect your agent