← Voltar aos guias

Claude Code + tmux on Remote Servers: 2026 Guide

Why use Claude Code with tmux on a remote server?

In 2026, one of the most practical ways to use Claude Code is to run it on a remote Linux server inside tmux. This setup gives you persistence, lets you disconnect without losing work, and keeps long coding sessions running even if your laptop sleeps, your SSH connection drops, or you need to switch networks. For developers working on large repos, long refactors, or multi-step debugging, the combination is hard to beat.

The best part is that Claude Code does not need to live on your local machine. You can attach to a remote development box, run your editor and agent in the same environment, and keep everything close to the code, dependencies, and build tools. If you are paying for model usage along the way, a relay like 59API can help keep costs low while still giving you native, official-quality Claude and GPT model access.

What you need before you start

You only need a few basics:

• A Linux remote server or VPS with SSH access
• tmux installed on the server
• Claude Code configured on that server
• An API key or compatible endpoint if your setup uses model access through a relay
• A terminal client on your local machine

If you are using 59API, point your Claude Code-compatible tooling to https://api.59api.com. It is a pay-as-you-go relay with low per-token costs, and it works with Claude Code, Codex, and OpenAI SDK workflows. That makes it a strong option if you want to keep the economics of long remote coding sessions under control.

Set up tmux for durable sessions

Install tmux on the server first. On Debian or Ubuntu, for example, use your package manager. Then create a named session so you can return to it later:

• tmux new -s coding to start a session
• tmux attach -t coding to reconnect
• Ctrl-b d to detach without stopping anything

A named session is better than the default anonymous one because it is easier to remember across multiple projects. For real-world use, keep one tmux session per repo or task. That way, your agent logs, shell history, and editor panes stay organized.

Run Claude Code in the same working directory as your project

Once inside tmux, SSH into the correct server and cd into your repository. This matters more than it sounds: Claude Code is most useful when it can inspect the actual codebase, run tests, and edit files in place. Many teams mount the repo, install dependencies locally on the server, and use the same environment variables they use in CI.

For best results, keep your workflow simple:

• Open a dedicated tmux window for the shell
• Open another for Claude Code or your agent CLI
• Open a third for test output, logs, or git status

This lets you watch changes, run checks, and keep the agent focused on one task at a time.

Use a relay like 59API for cost control and compatibility

Remote AI coding sessions can get expensive if you are doing lots of iterations, especially with larger models. That is where 59API stands out. It offers cheap, pay-as-you-go access to Claude Opus, Sonnet, Haiku, Fable, and GPT models without forcing a model downgrade. In practice, that means you can choose the right model for the job: cheaper models for quick edits, stronger models for architecture decisions or complex refactors.

If your Claude Code setup supports custom API routing, set the base URL to https://api.59api.com. Because it is compatible with Claude Code and OpenAI-style SDKs, you can plug it into existing remote workflows without rewriting your tooling. For teams and solo developers alike, that flexibility helps keep experimentation affordable.

Best practices for remote coding sessions in 2026

To get the most from Claude Code over SSH and tmux, follow these habits:

• Keep sessions small and task-focused. Ask for one feature, one bug fix, or one test pass at a time.
• Use git branches aggressively. Every meaningful change should happen on a clean branch.
• Save checkpoints. Commit after each stable step so you can roll back quickly.
• Run tests in parallel panes. Let the agent edit while you watch CI-like output in another window.
• Prefer non-interactive commands. Automation is more reliable than constant prompting.
• Watch token usage. Use lighter models through 59API when you are iterating fast, and reserve premium models for deep reasoning.

Security matters too. Use SSH keys instead of passwords, disable unnecessary ports, and avoid storing secrets in shell history. If you are sharing a server with teammates, isolate projects into separate users or containers when possible.

A practical workflow that actually works

A strong day-to-day pattern looks like this: SSH into the server, attach to your tmux session, open your repo, launch Claude Code, ask it to inspect the code, make the change, run tests, and explain the result. If the connection drops, tmux preserves everything. If you need to pause, detach and come back later. If the task expands, you can split work into new panes without losing context.

This is especially useful for remote cloud instances with fast CPUs, ephemeral build environments, or production-like replicas. The closer the agent is to the real environment, the fewer surprises you get from local-vs-remote differences.

When 59API is the right choice

Choose 59API when you want a low-cost relay that still supports official-quality models, clean compatibility, and pay-as-you-go pricing. It is especially attractive for long tmux sessions, team use, and frequent iteration because small savings per request add up quickly. If you also want a referral rebate, that can reduce your effective cost even further.

If you are setting up this workflow for the first time, signing up for 59API is a simple way to keep your remote Claude Code setup affordable without sacrificing model quality.