Run Cursor Agent CLI from your phone

The thing to know about Cursor before anything else is that the binary is not the one you installed for the editor. Happier looks for cursor-agent, the headless CLI, and deliberately never for cursor, which on most computers launches an editor window — no use on a computer nobody is sitting at. Cursor’s own CLI documentation writes the command as agent, so Happier will take a binary by that name too, but only after asking it about --format json and reading a Cursor version back: a stray agent on your PATH does not get to become your coding agent by accident.

Anysphere describes the CLI as a way to “interact with AI agents directly from your terminal to write, review, and modify code”, and puts an interactive interface and print automation for scripts and CI pipelines on the same footing. That second half is the reason this page exists — the thing runs perfectly well with nobody watching it, which makes the question of where you read it from a real question. Happier can also answer whether Cursor is signed in on that computer rather than shrugging: the same about call reports the account, and a CURSOR_API_KEY in the environment is honoured if that is how the box is set up.

Two Cursor behaviours needed work of their own on the Happier side. It generates images, and they arrive in the transcript as session media instead of a file path you have to go and find. And it asks structured questions — with choices, multiple selection and a freeform box — which the app draws as a question rather than as a wall of text you have to answer in prose; its plan step comes through as a plan you approve or send back. Model choice is read off your own CLI, and a model that advertises variants arrives as one entry with a reasoning-effort control and a fast toggle. Resume is the one place Happier is deliberately pessimistic: the release marks it experimental, checks it against the binary you have before offering it, and a resume that fails fails, rather than quietly opening a fresh conversation under the old title.

Cursor’s command-line agent is a separate binary from the editor: cursor-agent, which drives a repository from a terminal and therefore runs perfectly well on a box with no GUI. happier cursor starts it under your own Cursor account.

If the reason you are here is "I pay for Cursor and I am not at my desk", that is the whole feature. The agent runs where your repository is; you read it, approve it and steer it from whichever device you are holding.

The iOS and Android apps, the desktop app for macOS, Windows and Linux, and app.happier.dev are all clients onto the Cursor Agent CLI process running on your computer — not read-only mirrors. Answer a permission request, send the next instruction, browse the repository, read the diff. The transcript is end-to-end encrypted before it leaves your computer, so the server carrying it holds ciphertext.

Happier installs on the computer that already holds your repository, and happier cursor starts Cursor Agent CLI there as an ordinary subprocess, in the directory you point it at, signed in the way you signed it in. There is no gateway between the agent and Anysphere, and no copy of your source anywhere else.

Happier looks for cursor-agent on your PATH and runs the copy you installed. Anysphere distributes it with an install script rather than a package; Happier can show you that command, and it will not execute a vendor install script on its own — vendor recipes are refused unless you explicitly allow them.

Anysphere’s own install and sign-in guide is at cursor.com/docs/cli/installation. Happier does not mirror it, because a copy of someone else’s install instructions is a copy that goes stale.

Cursor Agent CLI is one of 13 command-line coding agents Happier runs, and they share one session list, one permission inbox, one set of keyboard shortcuts and one MCP configuration.

You define an MCP server once, and because Cursor Agent CLI does not consume MCP directly, Happier hands the same tool surface to it through the happier tools shell bridge instead — the same tools, delivered as commands it can run. The app previews the effective tool surface before a session starts, so which one you got is visible rather than inferred.

That matters most on the days you are running more than one: sessions from different vendors side by side, and a notification tap opening the session that raised the request rather than the app’s front door. The full list is here.

Typing happier cursor in a terminal starts the session right where you are standing, and the same session is in the app at the same time. What you read in that shell is Happier’s own display of the session — the transcript, the permission prompts, the tool output — rather than Cursor Agent CLI’s interface. Nobody has to give up the terminal to get a phone client.

There is no hand-off to Cursor Agent CLI’s own TUI. That exists for Claude Code, Codex and OpenCode, and for Cursor Agent CLI the session stays Happier’s on both ends. happier attach <session-id> lists the sessions on that computer and marks the ones it cannot reattach, rather than failing after you pick one.

The attach guide lists the cases instead of promising all of them.

One command on the computer that holds your code, then happier cursor in the repository you want to work in.

Walkthroughs are in the guides. All of it is MIT-licensed, and the relay is one you can run yourself.