Run coding agents from your phone, without SSH
The problem with a coding agent is not that it is slow. It is that it stops, and you are not at the desk.
Short answer: You can run Claude Code, Codex and other coding agents from your phone by letting the agent reach you: it pushes its question to your phone and waits for your tap. For the times you need to see the machine, a live screen and trackpad on the phone beats an SSH client on a tiny keyboard.
You set Claude Code or Codex going on something that will take a while, and you get up. Twenty minutes later it is sitting there waiting, because it hit one file it wants permission to overwrite. It has been idle the whole time you were making coffee.
That is the real cost of agentic coding, and it is not compute. It is the gap between the moment the agent needs you and the moment you notice.
How to run a coding agent from your phone
Why SSH is the wrong answer
The obvious fix is to get back to the terminal from wherever you are: SSH from a phone client, or a remote desktop into your own laptop. People do this and it works, in the sense that a bicycle works underwater.
- You have to notice something is waiting. Nothing tells you.
- A terminal on a phone screen means a tiny keyboard and no accurate cursor.
- Your laptop has to be awake and reachable, which on a closed MacBook it usually is not.
- You are exposing a shell to answer what is, most of the time, a yes-or-no question.
The mismatch is that you do not need a terminal. You need to answer one question and get out.
What the phone is actually good for
Three jobs, in the order they matter:
- Telling you it needs you. A push the moment the agent stops, not a tab you have to keep checking.
- Answering. Approve a file change, pick one of the options it offered, or type a one-line correction - ideally from the lock screen, without opening anything.
- Steering. Reading what it did, and occasionally driving the Mac properly, which needs a real pointer rather than a keyboard.
Only the third one needs anything resembling remote control. The first two are notifications and a reply box, which is why an SSH client is the wrong shape for almost all of it.
How the pieces fit together
The mechanism that makes this work is boring and worth understanding, because it is what decides whether your code ever leaves your machine.
A coding agent already speaks a protocol for calling out to tools - MCP. So the agent asks a question by calling a tool, that tool is served by an app running on your Mac, and the Mac sends your phone a push with a short preview of the question, so you can read it on the lock screen. Anything longer, like the full options or a screenshot, comes from your Mac when you open it.
That last detail is the one to look for in any tool that does this: whether the question and your answer travel through somebody's server, or only between your two devices. The push has to go through Apple - that is how iOS works - but nothing says the contents have to go with it.
And when you do need a pointer
Occasionally answering is not enough and you want to actually drive. A phone makes a decent trackpad for that: it is a slab of glass built to track a finger precisely, which is exactly the thing an SSH client cannot give you. MightyMouse turns your iPhone into a trackpad and microphone for your Mac, so the same phone that answered the agent's question can also move the cursor and let you talk to it.
Answer your coding agent from the lock screen.
Your iPhone as a trackpad and mic for your Mac. Free to try.
Common questions
Does my code get uploaded anywhere?
It depends on the tool, and it is the first thing to check. A push notification has to travel through Apple, but the question text and your answer do not have to - they can be fetched from your Mac directly over your own network.
Does this work when my laptop lid is closed?
No. The agent is running on your Mac, so the Mac has to be awake for it to be running at all. This solves being away from the desk, not being away from the machine.
Which coding agents does this work with?
Any that speak MCP, which is most of them now - Claude Code, Codex, Cursor and VS Code among others. The agent calls a tool; it does not need to know what is on the other end of it.
If your agent keeps stopping while you are away from the desk: MightyMouse puts your Mac on your phone - answer from the lock screen, and use the same phone as a trackpad when you need to drive.