// AGENT INBOX · A CLIZE PRODUCT
A real email inbox for your AI agent
Send with your approval. Receive as untrusted data. Pass email verification on its own. One install, inside Claude Code and Codex.
Agent Inbox is an email API for AI agents: a real inbox the agent can send from and receive into, with a human-approval gate on outbound, untrusted-by-default inbound, and email-verification handling built in. It is the email product of Clize — Clize is a CLI and MCP server that gives AI coding agents real-world actions: domains, email, deploys, payments, and media generation. Install once and your agent has an address of its own, reachable and able to act under its own name, without your personal mailbox ever being in the loop.
What an email API for AI agents actually needs
Transactional email APIs push one-way notifications for an app. An agent needs more: an address that belongs to it, an inbox where replies and verification codes land, threads that survive across sessions, and hard lines on what it can do unattended. That combination — send, receive, continuity, gates — is the category. The comparison of the current options shows how the tools split across it.
Send — with a human okay
The agent drafts. You approve. Then it sends, under your name. That order is enforced by the platform, not by a prompt you hope the model follows. Nothing goes out unattended, so the worst an over-eager agent can do is show you a draft. See the flow in let your agent send an email.
Receive — untrusted by default
Everything that lands in the inbox — a customer reply, a signup confirmation, a stranger's message — is parsed and handed to the agent as data to read, never as instructions to obey. A hostile email cannot redirect the agent, because inbound content has no authority. That single default closes the most common prompt-injection path an agent with email faces.
Pass email verification on its own
When your agent signs up for a service, the "verify your email" step usually ends the run. With a real inbox it doesn't: the verification mail lands, the agent reads the code with clize email inbox --wait-for, and the signup completes. Light identity gates, cleared without you; heavy gates — money, or a real person's identity — stay yours. Walkthrough: AI agent email verification.
Transactional sending for your app
The same platform sends on behalf of your server, not just your agent: a scoped key (clize_sk_, send-only, limited to your domains) with idempotency, signed delivery webhooks, and RFC 8058 one-click unsubscribe. Your product's digests and your agent's conversations share one sending reputation you control.
Works where your agent works
Agent Inbox rides the Clize CLI and MCP server, so it installs into Claude Code and Codex in one step. The CLI's output is written for an agent to read; the MCP server exposes the same actions as structured tools for hosts that want them.
Part of Clize
An inbox rarely acts alone. The address lives on a domain the agent can register, next to a site it can deploy and payments it can collect — one account, one set of safety gates. Start with email and the rest of Clize's real-world actions is already wired. For the thinking behind giving an agent its own address at all, read why an AI agent needs its own email.
FAQ
What is Agent Inbox?
Agent Inbox is the email product of Clize: a real inbox your AI agent can send from and receive into. Outbound mail is drafted by the agent and sent only after a human approves; inbound mail is parsed as untrusted data; verification codes can be read to finish signups. It installs into Claude Code and Codex as a CLI and MCP server.
What is an email API for AI agents?
An email API for AI agents gives an agent its own address and programmatic send and receive — unlike transactional email APIs, which only push one-way notifications for an app. The agent can hold a thread: receive a reply, read a verification code, draft a response, and continue across sessions.
How is Agent Inbox different from AgentMail?
AgentMail is an API-first inbox platform you call from your own code. Agent Inbox is coding-agent native: it installs into Claude Code and Codex as a CLI and MCP server, ships with a human-approval gate on outbound and untrusted-by-default inbound, and shares one platform with domains, deploys, and payments. If you are building an agent product on an inbox API, AgentMail is a strong choice; if your agent lives in a coding agent and you want gates by default, that is what Agent Inbox is for.
Can my agent send email without me?
No. The agent drafts; you approve; then it sends under your name. That outbound gate is the default and not a setting you have to remember to turn on.
Can the agent receive verification codes?
Yes. The inbox is real, so when the agent signs up for a service, the verification mail lands there and the agent reads the code to continue — light identity gates, cleared on its own. Heavy gates tied to money or a real person's identity stay with you.
Does it work with Claude Code and Codex?
Yes. One install wires Clize into Claude Code and Codex — as a skill by default, or an MCP server if you prefer structured tools. The same inbox works from either.
Can my app send transactional email through it?
Yes. A scoped sending key (send-only, domain-limited) lets your server send digests and notifications through the same platform, with idempotency, delivery webhooks, and one-click unsubscribe handling built in.
Give your agent an inbox.
A free handle includes a working address the agent can send from and receive into — approval on outbound, untrusted inbound, verification codes readable.
$ npm i -g @clize/clize $ clize login $ clize claim # your agent's address, live[ Compare the options → ]