AI access methods
Three ways — none require manually copying secrets. Authorize once in the browser.
Unified entry · let AI choose capabilities
When you are not sure whether to use deploy, user auth, or a future capability, send:
Choose and use PocketBay capabilities based on my needs. Capability directory: https://pocketbay.com/ai
AI will query the capability directory first, then read the specific runbook. Directory: /ai。
Method 1 · one-line (zero config, recommended)
Send this to your AI coding agent (Cursor / Claude Code / Codex, etc.):
Please deploy the current project to PocketBay. Deployment guide: https://pocketbay.com/deploy
AI fetches the link, reads the deployment runbook, creates a pairing session, and gives you an authorization link. After you approve, it packages, uploads, and builds; on failure it reads logs and repair guidance, updates code with your consent, and retries up to 3 times.
Visual guide: /deploy;machine-readable runbook: /api/deploy/instructions。
One-line PocketBay user auth
When your app needs login or per-user data, send this to AI:
Add PocketBay user auth to the current project. Integration guide: https://pocketbay.com/app-auth
AI picks the correct identity API for static or dynamic projects, updates data isolation, and reminds you to redeploy and enable project auth.
Visual entry: /app-auth。
Method 2 · CLI
pip install pocketbay cd your-project # framework is chosen by AI after analysis, e.g. nextjs: pocketbay deploy --framework nextjs
First run prints an authorization link — click once (same directory is authorized later). On success you get the live URL. AI can also invoke it directly: failures exit non-zero with build logs so AI can fix and rerun.
Method 3 · MCP Server
After install, add one stdio server in your AI client's MCP config:
pip install "pocketbay[mcp]"
{
"mcpServers": {
"pocketbay": {
"command": "pocketbay-mcp"
}
}
}Tools: start_deploy → check_authorized → finish_deploy → get_status (failure self-heal loop).
Deployment conventions
- · Supported types: static / node / nextjs / python / dockerfile — framework_hint must be passed after local AI analysis; the platform does not auto-detect
- · Apps must listen on
PORT(default 8080) and bind0.0.0.0 - · Projects with
index.htmlare treated as static; projects with aDockerfilebuild from it - · Live URL after deploy:
https://<project>.pocketbay.app