Marina has an MCP server, a command line built to be driven, and an API. Connect a coding agent in one line and it ships your app, attaches a database and reads the logs, spending from a prepaid wallet instead of a card.
No extra charge
Point your client at the server and it signs in through the browser, the way the site does. There is no key to paste into a config file you might commit.
The wallet is prepaid, so it can only spend what you topped up. Give it a credential that cannot spend at all, or one with a limit. Revoke it and every session using it stops.
Every command speaks JSON, none of them ever prompts, and the exit codes mean something. An agent finishes unattended instead of stalling on a keystroke.
Creating, deploying and resizing go ahead on their own. Buying a domain is permanent, so that one comes back to you first.
$ claude mcp add --transport http marina https://mcp.marinahost.app/mcp
Then, in the client:
deploy this repo and give me the URL
Building…
Live · shop.sites.marinahost.app
Cursor and Codex read the same server, and anything that speaks the protocol can use it. If you would rather it went through a terminal, the command line does everything the server does, and the API is underneath both.
Read the setup →The whole product, not a read-only slice of it: everything your agent can reach through the server.
Create an app, deploy it from a repo or a directory, publish a static site, set config vars, resize, run a one-off command inside it.
Create Postgres, Redis or SQLite, attach one to an app, and read the connection back.
Search for a domain, buy it, point it at what you run, and set up email on it.
Read the build log, read the runtime log, and ask what any of it costs before spending.