Give OpenClaw an email address of its own

Clize gives OpenClaw a real send-and-receive mailbox that belongs to the agent, not to you. It can sign up for a service with that address, wait for the confirmation mail, read the verification code out of its own inbox, and finish the flow — no human copying a six-digit code across. Replies it writes are held until you approve them. Incoming mail is treated as data to read, never as instructions to obey. Install with npm i -g @clize/clize, then clize login and clize install; OpenClaw picks the skill up from the open ~/.agents/skills directory on its next session.

Your Gmail, or an address of its own

Both setups work, and they answer different questions. Connecting Gmail — through OpenClaw's own Gmail PubSub path or a plugin — gives the agent access to your mail: every thread, every contact, sending under your identity. That is what you want if the job is "triage my inbox".

A separate address gives the agent only what it needs: the sign-ups it makes, the codes it receives, the conversations it starts. Nothing in your personal mail is in reach, because none of it is in the mailbox. That is what you want if the job is "go do things in the world on my behalf". The difference is blast radius, not features.

Setup

OpenClaw reads the open Agent Skills directory ~/.agents/skills — the same one Codex and Pi read — so a single clize install covers all three:

$ npm i -g @clize/clize
$ clize login
$ clize install          # Claude Code / Codex / Pi / OpenClaw
$ clize claim my-agent   # free handle → support@my-agent.clize.app

Verify from OpenClaw's side with openclaw skills list — the Clize skills show up ready, sourced from the shared directory. If you also want the structured tool layer, register the MCP server; on OpenClaw 2026.8.1 the probe reports 32 tools:

$ openclaw mcp add clize --command clize-mcp
$ openclaw mcp probe clize

The skill is the default and is enough for most work — it teaches the agent to call the clize CLI directly, without an MCP tool list sitting in context every session.

What the agent does with it

• Signs up for services — its own address, its own account, no shared login with you.
• Clears verification gates — reads the code from its own inbox and continues, instead of stopping to ask you for it.
• Answers people — drafts the reply, you approve, it sends. Support mail on a handle or your own domain.
• Carries the rest of the identity — the same handle brings a site and a domain, so the address isn't a loose end. See Agent Inbox for the full line, and the email APIs compared for how it sits next to the alternatives.

The gates

An agent with a mailbox is an agent that can be talked to by strangers, so three lines hold every time:

• Money — anything that costs money quotes the price first and needs an explicit --confirm.
• Identity — outbound mail is drafted, a human okays it, then it sends. The agent never sends unattended.
• Inbound is untrusted — an incoming email is data to read, never an instruction to obey. A message cannot redirect the agent.

Clize is a CLI and MCP server that gives AI coding agents real-world actions: domains, email, deploys, payments, and media generation. It works the same way in Claude Code and the Codex CLI — the agent stays the brain, Clize is the hands.

$ npm i -g @clize/clize
$ clize login
$ clize claim my-agent