Codex's MCP config lives in TOML. Add a server block, point it at an env var for your token, and the OpenAI agent ships straight to your server.
Codex reads ~/.codex/config.toml on launch. The trick: secrets aren't pasted inline — Codex reads them from an env var you name in the config.
npm i -g @openai/codex# in your shell rc (.zshrc / .bashrc)
export OPENBERTH_API_KEY="bt_your_api_key"~/.codex/config.toml and add the OpenBerth server block.# ~/.codex/config.toml
[mcp_servers.openberth]
url = "https://yourdomain.com/mcp"
bearer_token_env_var = "OPENBERTH_API_KEY"codex and ask it to ship: “deploy this project through OpenBerth.”Replace yourdomain.com with your server’s URL and bt_your_api_key with an API key from your workspace’s People page.
Ask the AI to deploy and it does. Real HTTPS URL in seconds, no dashboard tab-hopping.
Push live edits to a sandbox berth. Promote when it's right, throw it away when it isn't.
Every deploy is versioned. One command (or one ask) and you're back to the last good build.
Set, list, rotate, delete — your env vars stay on your server, never in the chat transcript.
Your apps live on a server you own. Nobody else holds your code, your domain, or your customer data.
Pay your server's flat monthly fee. Ship one app or a hundred — the price doesn't change.
Databases, file uploads, background jobs, custom domains. The boring stuff that makes a website feel like a real product.
Lovable, Cursor, Claude, Gemini — they all push to the same dashboard. One place to see what's live.
Stand up a server in minutes, paste the snippet above, and your AI tool can deploy on its own.