Run Claude Code and Codex from your phone: Happy CLI or Remote Control?

You can control Claude Code from a phone with Anthropic’s Remote Control, or use the original Happy CLI to reach Claude Code and Codex through the same mobile app. The agent still runs on your computer. Keep that host awake and check its subscription login before leaving your desk.

Originally checked 27 October 2026; revised and source-checked 2 November 2026. This is a local-session setup guide, not a claim that a phone runs your development environment. The Happy listing leads to its maker; the Claude and ChatGPT hubs cover other subscription-compatible apps.

Which phone-control route should you choose?

Start with the runtime you already use. Anthropic’s Remote Control documentation covers Claude Code. Happy’s terminal guide documents wrappers for both CLIs.

Your starting point	Suggested route	First check
One Claude Code session	Native Remote Control	Eligible Claude login and organization permission
Claude Code and Codex in a terminal	Original Happy CLI	Each underlying CLI’s active account
Happy Desktop already installed	Its Mobile Access setup	Follow Desktop instructions, not the CLI pairing guide
Laptop must be switched off	A separately evaluated remote host	Where execution and credentials will live

The CLI/Desktop distinction matters. Happy’s maker now describes the original CLI as maintenance-mode software, with critical fixes continuing and new features going to Desktop. This article deliberately covers the CLI wrapper for existing terminal users. Desktop and Happy Agent are separate components; do not assume their architecture or billing route is identical.

How do you connect Claude Code with native Remote Control?

In the project’s existing Claude Code session, run /remote-control. Alternatively, start server mode from a terminal with claude remote-control. Use the displayed session link or QR code to connect from the Claude mobile app or claude.ai/code, following Anthropic’s current steps.

The documentation lists Pro, Max, Ultra, Team and Enterprise support, excludes API-key authentication and requires organization enablement for Team and Enterprise. Sign in through Claude’s own flow. If access is denied by organization policy, ask the administrator rather than substituting a proxy or copying credentials.

Our suggested first task is read-only: ask for the name of the current branch and a summary of one non-sensitive file. Confirm the response appears in both views. Only then try a small change on a disposable branch. This is a proposed acceptance test, not a test result from our machines.

How do you connect Claude Code and Codex through Happy CLI?

First make sure the underlying Claude Code or Codex installation works locally. Then follow Happy’s maker-maintained README and terminal guide: install the npm package with npm install -g happy, and start the desired runtime with happy claude or happy codex.

The maker documents QR pairing on first use. Use the Happy phone app to complete that flow. For Desktop, the mobile guide instead directs you to Settings → Mobile Access; do not mix those instructions with a CLI troubleshooting session.

For plan-funded work, check the agent login separately from the Happy pairing. Anthropic’s plan help page warns that ANTHROPIC_API_KEY can select API billing instead of subscription usage. OpenAI’s authentication guide distinguishes ChatGPT sign-in from API-key billing. A successful phone connection proves neither account choice nor remaining allowance.

What should you test before leaving your desk?

Use a short handoff checklist rather than discovering a missing permission on a train:

1. Host: identify the computer doing the work and keep it powered and connected.
2. Control: send a harmless follow-up from the phone and verify the terminal receives it.
3. Approval: confirm you can inspect and answer a permission request without granting blanket access.
4. Stop: check how to stop the current turn before starting a long task.
5. Review: leave commits, production changes and deployments for a deliberate review.

Happy’s mobile page lists permission replies and stopping a turn among its controls. Those controls do not establish that an action is safe. Keep repository secrets out of prompts and give the agent only the access the task needs.

Why does phone control stop working, and does it change your limits?

For native Remote Control, Anthropic says local execution requires the host and Claude process to keep running. Start diagnosis there, then inspect login, connection and organization-policy errors. Do not treat a disconnected view as proof that the local task finished or stopped.

For Happy CLI, verify which machine and session you paired, then use the maker’s terminal documentation. Avoid deleting authentication state as a generic repair. Happy’s terminal guide explicitly warns against deleting its shared state directory.

Neither a phone nor another control surface adds model allowance. Claude’s plan help page says Claude and Claude Code share limits. Before starting more work remotely, inspect the active account and usage; for the broader billing distinction, use our subscription setup guide.