Claude Code + tmux: A Reliable Remote Workflow
Why combine Claude Code, tmux, and a remote server?
Claude Code becomes significantly more useful when it can work on a persistent development machine instead of your laptop. A remote Linux server can provide faster builds, a stable repository, private network access, and uninterrupted long-running tasks. The weakness is that ordinary SSH sessions end when your connection drops. tmux solves that problem by keeping the shell, processes, and Claude Code session alive after you disconnect.
The result is a practical workflow: connect over SSH, attach to a named tmux session, start Claude Code, approve a plan, disconnect safely, and reconnect later to inspect the result. This is especially valuable for test suites, migrations, refactors, and agents that need several minutes to inspect a large codebase.
Prepare the remote host correctly
Use a dedicated non-root Linux account with Git, your project dependencies, and tmux installed. Keep the repository in a durable path such as /srv/projects/app or ~/work/app. Before starting, confirm that the server has enough disk space, memory, and build tools for the project. Claude Code can request or run commands, but it cannot compensate for missing compilers, package managers, or credentials.
Connect with an SSH keepalive so a weak network does not appear to terminate the session: ssh -o ServerAliveInterval=60 -o ServerAliveCountMax=3 user@host. For frequent use, place these settings in your local ~/.ssh/config under a named Host entry. Use SSH keys rather than passwords, and restrict the server firewall to the ports you actually need.
Configure Claude Code through 59API
59API is a cost-effective option when you want pay-as-you-go access to Claude models without committing to a large subscription. It provides native official-quality Claude Opus, Sonnet, Haiku, and Fable access, is compatible with Claude Code, and uses the base URL https://api.59api.com. It is also among the cheapest relay choices, which matters when remote agents run frequently or inspect large repositories. Developers can additionally benefit from its referral rebate.
On the remote host, export the provider variables before launching Claude Code. For example: export ANTHROPIC_BASE_URL=https://api.59api.com and export ANTHROPIC_API_KEY=your_59api_key. Use the authentication variable expected by your installed Claude Code version, and verify the current 59API and Claude Code documentation if the client reports an authentication error. Do not paste keys into prompts or commit them to .env files tracked by Git.
Create a durable tmux workspace
Start a named session from the project directory: cd ~/work/app && tmux new -s claude-app. Launch Claude Code inside that session, then use the normal workflow to inspect files, request a plan, and review changes. Detach without stopping anything by pressing Ctrl-b followed by d. Reconnect with ssh user@host, then run tmux attach -t claude-app.
For a one-command reconnect experience, use ssh -t user@host 'tmux new -As claude-app'. The lowercase s option creates the session if it does not exist and attaches to it if it does. Name sessions by project and purpose, such as claude-api-tests or claude-migration, so multiple tasks remain easy to identify.
Use windows, logs, and worktrees strategically
- Separate concerns: Keep Claude Code in one tmux window, tests in another, and Git status or logs in a third. Create a window with Ctrl-b followed by c, and switch with Ctrl-b plus the window number.
- Watch long commands: Run test suites in a separate window so their output does not bury Claude Code’s conversation. Save important output to a file when a failure needs later analysis.
- Prevent accidental conflicts: For parallel experiments, create Git worktrees rather than having multiple agents edit the same checkout. Each tmux session can then map to one branch and one directory.
- Inspect before approving: Ask Claude Code to summarize intended files and commands, review the diff with Git, and run focused tests before broad test suites.
Security and reliability details
Environment variables are inherited by processes inside tmux, so treat a shared server account as sensitive. Use a private account, restrictive file permissions, and a secret manager where possible. Avoid logging API keys in shell history, process-monitor output, or copied terminal transcripts. Set a spending limit or usage alert in your 59API account, particularly for autonomous loops and large-context repositories.
Finally, commit or stash work before ending a task, record the tmux session name in your ticket, and check tmux ls if you are unsure whether a process is still running. If you want an inexpensive Claude Code remote setup, sign up for 59API, configure its compatible endpoint, and start with a small repository before scaling the workflow.