for people who prompt their code

The missing deploy button for your AI IDE.

Claude, Cursor, Codex, Windsurf — they all write the code. Cubby is the thing that takes it from cursor blink to a URL your friends can actually open.

⎇ claude-code · MCP ready↻ cursor.rules + CUBBY.md✶ codex-friendly
claude-code · cursor · codex
⟨/⟩

CUBBY.md in every starter

A context file your AI agent reads first. Deploy rules, auth helpers, db conventions — it already knows.

MCP

Ship from Claude Code

deploy_app, get_logs, get_app, and 16 more tools exposed over MCP. Your agent ships without switching terminals.

Production mirror, locally

cubby dev spins up the exact prod stack. Your agent iterates against the same DB and auth it will deploy to.

Secrets via CLI, not .env

cubby secrets set OPENAI_KEY --env prod — NAME only, it prompts for the value. Never commit a key, never hand your model a .env.

Snapshot before, restore after

cubby db snapshot, then let the agent try its idea. cubby db restore if it goes sideways.

Logs that an agent can read

cubby logs --json pipes structured output. Give your agent eyes on production.

Your agent wrote it. Ship it.

Install the CLI, add the MCP server, deploy by the end of this paragraph.

Claim your cubby →