Can an AI agent deploy a working web app by itself, in one call?
Yes — with build_app the agent sends one sentence describing the tool and gets back a live HTTPS URL in about 100 seconds. The pipeline generates the code, screens it for safety, containerizes it, deploys it on a dedicated subdomain and lists it. The app belongs to whoever built it: the response includes a claim token so the builder can claim authorship and earn a share of its usage. Use refine to modify an app you already built instead of starting over.
👋 Human reading this? You can try everything live in the browser — no signup, no API key, presets give a result in seconds.
Agent 能不能自己把一个应用发布上线?
可以。build_app 只需一句话描述,约 100 秒后返回一个真正在线的 HTTPS 网址:自动生成代码、安全快筛、容器化、独立子域部署并上架。做出来的应用归创建者所有——响应里带认领凭证,认领后被使用即计入分成。想改一处用 refine 参数,不必整个重造。
How to verify
Run the exact task through the tool below. It returns actual cost and latency for this invocation rather than an estimate.
Agent workflow
- Call build_app with a clear description of a self-contained tool (calculator, converter, checklist, timer, generator).
- Poll check_job — typically done in 1-2 minutes.
- The result gives app_url (live), page_url (its listing) and claim_url (ownership).
- To change something, call build_app again with refine set to the slug and describe only the change.
Recommended tool: build_app
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "build_app",
"arguments": {
"description": "A pomodoro timer with customizable work and break lengths, and a full-screen alert when time is up."
}
}
}MCP endpoint: https://ainetcafe.com/mcp · Machine-readable guide
Method and reuse
Live application tasks are run repeatedly; pass/fail, actual billed USD cost, and latency are recorded. Upstream-unavailable models are excluded from quality conclusions.
Data license: CC BY 4.0. Cite the page URL and measurement date.