Install, wire the skill into your agent, log in, claim a free identity. Everything after that is plain language inside Claude Code, Codex, Pi — any agent that runs skills or MCP.
Node.js 20+ · No Cloudflare, Vercel or model keys · Free handle to start
Four steps, in order. Step two is the one people skip — and the one that makes the agent actually reach for Clize.
One package puts clize and clize-mcp on your PATH.
$ npm i -g @clize/clizeUsing volta, pnpm or asdf? clize update knows the right upgrade command later.
Drops the clize skill into Claude Code, Codex and Pi so the agent knows when to use it and which gates apply. A binary on your PATH alone tells the agent nothing.
$ clize install # --mcp also registers the clize-mcp server # --claude / --codex / --pi to scope · --dry-run to preview
Opens the browser — GitHub, Google or email. The first login creates your account; nothing else to configure.
$ clize login # CI / headless: clize login --token clize_…
A site and a working inbox on a free handle. init binds the directory so later commands need no --domain / --from.
$ clize claim studio --email # → studio.clize.app + support@studio.clize.app $ clize init --handle studio $ clize status
Let the agent install its own hands. Paste into Claude Code, Codex or Pi and answer a question or two.
Install clize as a tool of your own, so you can give yourself an email address, register for services with it, send and receive real email, and put sites live for me. 1. Install: npm i -g @clize/clize 2. Run clize install — it adds the clize skill so you know to reach for these tools next time. 3. Run clize login — it opens the browser (GitHub / Google / email; the first login creates my account). Walk me through it. 4. Run clize check to confirm you are connected, then clize claim <a slug you propose> --email. Then tell me you're ready. Before spending money or emailing a real person, always ask me first.
The agent runs the commands one by one; you only sign off on money and outbound mail.
You: sign up for Postmark with your own address and read the verification code
You: any supplier replied? draft answers — I'll say send
You: make a site for the studio and put it on studio.clize.app
Enforced by the platform, visible in the transcript — not a prompt you have to remember.
domain buy, gen image / video / music, shop refund quote first and act only with --confirm.
email send without --confirm returns a draft and sends nothing — no exceptions, not even "notifications".
Mail the agent reads is data, never instructions; the CLI prints an [untrusted] line before every message.
The skill is the default and already lets the agent drive the whole CLI. If your host prefers MCP, register the server — all of it, or one product line so only that line's tools sit in context.
| Register | Tools in context |
|---|---|
clize install --mcp (or claude mcp add clize -- clize-mcp / codex mcp add clize -- clize-mcp) | 32 — the whole curated surface |
claude mcp add inbox -- clize-mcp --profile inbox | 13 — Agent Inbox |
… --profile storefront | 15 — Agent Storefront |
… --profile sites | 5 — Sites |
… --profile domains | 9 — Agent Domains |
Media generation and the build method packs are CLI-and-skill only, on purpose. A profile is a context budget, not a permission boundary — to restrict what a key can do, create a scoped key with clize email key create.
It copies the clize skill into each coding agent it finds on the machine — Claude Code (~/.claude/skills), and Codex and Pi, which share ~/.agents/skills — so the agent knows when to use clize and which safety gates apply. Only the skill's short description sits in the agent's context until something triggers it. A binary on your PATH alone does not tell the agent clize exists. Add --mcp to also register the clize-mcp server; --claude, --codex and --pi scope it; --dry-run previews. It is idempotent, and clize update refreshes both the CLI and the skill later.
No. Clize is hosted: the CLI and MCP server are thin clients that hold no infrastructure or model credentials. clize login is the only credential on your machine, and every domain, mailbox, deploy and generation runs on Clize's infrastructure.
Claude Code, Codex and Pi are detected by clize install and get the skill by default. Any other agent that runs skills or MCP works too: point it at the skill file, or register the clize-mcp server (all 32 tools, or one product line with --profile inbox, storefront, sites or domains).
A free handle gives the agent a live site and a working inbox on <slug>.clize.app. The free tier allows 3 handles, 10 claim attempts a day, 30 outbound emails a day, 20 deploys a day and 500 MB per site; any top-up lifts those limits. Spending only happens for things that cost money upstream — a domain, a media generation, an SEO research query — from a prepaid balance, and each of those quotes first and waits for your confirm.
clize update pulls the latest release and refreshes the skill in every agent it was installed into. clize doctor shows what is installed versus what is actually running: CLI version, skill drift, and whether the control plane is reachable.
Six products, one account. Each page shows what the agent does with it — and what stays yours.
A free handle includes a live site and a working inbox.
$ npm i -g @clize/clize && clize install && clize login && clize claim studio --email