an agent signed up before the humans did

Last week I was going through InstantDeploy's user table and found a signup from

. That's not a person. That's an autonomous agent on Coze — ByteDance's agent platform — that needed to put something on the web, found InstantDeploy, created an account, and deployed. Twice.

Nobody marketed to it. There was nothing to market with — we hadn't launched. An agent simply did what agents do: searched for a tool that solved its problem with the least friction, and used it.

That one row in the database is the entire thesis of this product.

What InstantDeploy is

One command, live URL, about two seconds:

curl -fsSL instantdeploy.site/install.sh | bash -s -- .

No account, no repo, no build pipeline, no dashboard. Anonymous deploys expire in 24 hours; claim one (or use an API key) and it's permanent. Every deployed site carries a small Remix badge — anyone viewing it can hand the full source to their agent and ship their own version.

What we shipped in the last sprint

The last stretch was the most productive this project has ever had. All of this is live right now:

• MCP server —
  instantdeploy-mcp
  gives Claude, Cursor, and any MCP client a native deploy tool. Your agent ships its work without leaving the chat.
• Custom domains with automatic SSL — add two DNS records, and your site is live on your own domain with a certificate issued on the first request. No provisioning queue, no dashboard wizard.
• Presigned uploads — archives go straight to storage, so big sites don't care how busy our API is.
• Named API keys — one key per agent or project, individually revocable.
• Unlisted sites — share a draft by link without it appearing in the public feed.
• Per-site analytics — every site now counts its views, visible in the dashboard.
• Real abuse defenses — report buttons on every deployed page, admin takedowns, disposable-email blocking, deliverability checks at signup, and email verification before anything becomes permanent.

What building for agents actually changes

Three things surprised me building this:

1. Your docs are an API. Agents don't read landing pages — they read

and . We treat those files as products now. If an agent can't figure out the entire flow from one fetch, that's a bug.

2. Trust systems can't rely on friction. Every classic anti-abuse tool — captchas, email confirmation walls, "click here to continue" — assumes a human you can inconvenience. Our answer: signups are instant, but permanence is gated. A fake email gets you exactly what anonymous access already gets you: nothing extra. The incentive to fake disappears instead of the door closing.

3. Distribution means being in the context window. The old playbook is SEO and ads. The new one is being installed in the toolset before the need arises — MCP registries, skill indexes, one-line installs. When the agent needs hosting, the tool that's already loaded wins.

What's next

Site Data (forms and polls for static sites), a proper Pro tier, and version rollbacks. The roadmap is public at instantdeploy.site/roadmap.

If you're building with Claude Code, Cursor, or any agent that can run a shell command — give it a deploy button and see what it ships:

npx skills add instantdeploy-site/instantdeploy.site --skill instantdeploy -g

— Ali (aliarain.com) — Desktop App Lead at CommandCode.ai, building InstantDeploy in public.