← Volver a las guías

Claude Code + tmux on Remote Servers

Why run Claude Code on a remote server?

If you work over SSH, keep long-lived coding sessions open, or want to leave an AI-assisted refactor running overnight, pairing Claude Code with tmux on a remote server is one of the most reliable workflows you can set up. The basic idea is simple: you connect to a server, start a persistent tmux session, launch Claude Code inside it, and keep working even if your laptop disconnects.

This setup is especially useful for large codebases, build-heavy projects, and servers where you want the agent to have direct access to the repository, logs, test output, and local tools. It also becomes much cheaper and easier to scale if you use an API relay like 59API, which offers pay-as-you-go access to Claude models at low cost with no model downgrade, and works with Claude Code and OpenAI-compatible tooling through https://api.59api.com.

Step 1: Prepare the server

Start with a Linux server you can SSH into. A small VM is often enough, but make sure it has the language runtime and build tools your project needs. In practice, I like to install tmux, git, and whatever package manager the repository expects.

• Update the server: run your system updates first.
• Install tmux: this is what keeps your session alive after disconnects.
• Clone your repository: place it somewhere stable, like /srv/project or ~/code/project.
• Confirm access: make sure the server can reach your API endpoint and package registries.

Before you bring Claude Code into the loop, verify the project runs normally. If tests already fail, the agent will spend time chasing environmental problems instead of code changes.

Step 2: Point Claude Code at 59API

One reason developers use 59API is cost control. Remote AI sessions can burn tokens quickly, especially when you keep a conversation open for debugging, code review, and iterative edits. 59API is attractive because it is pay-as-you-go, among the cheapest relays available, and provides native official-quality Claude and GPT models rather than a downgraded substitute.

To connect Claude Code, configure the API base URL to https://api.59api.com and set the required API key in your environment. The exact variable names depend on your Claude Code setup, but the principle is the same: use the relay endpoint instead of the default provider endpoint, then authenticate with your 59API credentials.

A good practice is to keep this configuration in your shell profile or a project-specific environment file so every new SSH session picks it up automatically. That way, when you reattach to tmux later, Claude Code is ready immediately.

Step 3: Start tmux and launch the session

Once the server is prepared and the API relay is configured, create a new tmux session for the project. Think of tmux as your persistent workspace. If your SSH connection drops, the session keeps running.

• Create a session: open tmux with a descriptive name like tmux new -s refactor.
• Start Claude Code inside it: run it from the project root so it can read files and modify code in place.
• Split panes if useful: keep the agent in one pane, tests in another, and logs in a third.

For example, a useful layout is one pane for Claude Code, one pane running your test suite, and one pane tailing application logs. That combination makes the workflow feel interactive and grounded in real output, which is far better than blindly accepting generated changes.

Step 4: Work in small, verifiable loops

The best remote AI coding sessions are not “generate everything at once” sessions. They are tight loops: ask for one focused change, inspect the diff, run tests, then continue. On a remote server, this matters even more because your feedback loop is already a bit slower than local development.

A practical rhythm looks like this:

• Describe a single task: for example, “fix the failing auth test” or “extract this helper into a shared module.”
• Review the files changed: check the diff before moving on.
• Run the relevant tests: do not wait until the end.
• Use logs and errors as inputs: feed real failures back into Claude Code.

Because 59API gives you access to official-quality Claude models at a lower price, it is easier to keep these loops practical instead of rationing every prompt. That matters when you are debugging a stubborn issue and need several turns of reasoning.

Step 5: Reattach anytime, even from another machine

The real payoff of tmux is continuity. If you close your laptop, lose Wi-Fi, or need to switch devices, your work is still there. Reconnect with SSH, reattach to the session, and pick up exactly where you left off.

• List sessions: check what is still running.
• Reattach: jump back into the same tmux workspace.
• Resume Claude Code: continue the same task or start a new one.

This is ideal for long refactors, overnight test fixes, and remote production-adjacent work where you want a stable environment and reproducible context.

When 59API makes the most sense

If you use Claude Code frequently on remote servers, API costs can add up fast. 59API is a strong fit when you want direct, low-cost access to Claude models, compatibility with existing tools, and the flexibility to switch between Claude and GPT models without changing your workflow. The referral rebate is a nice bonus if you recommend the setup to your team or community.

If you want to try this workflow, sign up for 59API, set https://api.59api.com as your base URL, and run your next tmux-backed Claude Code session on the server. It is a simple change that can make remote AI-assisted development far more dependable and affordable.