Claude Code, anywhere

Claude Code is Anthropic’s coding agent for the terminal. You start it in a project directory and it reads the codebase, edits files, runs commands, and asks before doing anything it has not been told it may do. It installs with one line on macOS, Linux, and WSL, or through Homebrew, WinGet, apt, or npm; it ships as a native binary that updates itself, and runs on macOS 13 or later, Windows 10, and recent Ubuntu, Debian, and Alpine with 4 GB of memory. The same engine sits inside the VS Code and JetBrains extensions, a desktop app, and Anthropic’s hosted Claude Code on the web.

Its behaviour is shaped by a few plain files. A CLAUDE.md at the root of the repository gives it standing instructions and is meant to be committed; a shared settings file holds a team’s permissions and hooks, with a local one for personal overrides; an MCP file adds tools. Sessions are saved per project directory and can be resumed after the terminal closes. What it is not is a chat window — the line between Claude and Claude Code, and the first hour of running it, each have a page in the rack below.

The software is free to install and does nothing without a paid route to the model; the free Claude.ai plan does not include it. Pro at $20 a month does, with an allowance measured in five-hour sessions and a weekly cap; Max at $100 or $200 multiplies that allowance five or twenty times and defaults to Opus; Team seats put the same thing under an admin at $25 or $125 a seat; Enterprise bills usage at API rates behind SSO and managed settings. Or skip the plans and pay per token with a Console API key, or through Amazon Bedrock, Google Cloud, or Microsoft Foundry. A usage command inside the tool shows what you have spent, and a status command shows which route is paying.

The choice is a real one — twenty days of average API-billed use costs more than Max 5x, but a subscription caps the usage as well as the bill — and it is the subject of three pages here: whether Claude Code is free, what the Max plan actually buys, and a plan calculator that does the arithmetic for your hours.

A coding agent runs for minutes or hours between the moments it needs you, so the question that follows install is not how to use it but how to reach it — from the couch, from the phone, from another machine, after the laptop closes. Anthropic answers part of that: Remote Control puts a local session in the Claude app, and Claude Code on the web runs sessions on Anthropic’s machines. The older answer, SSH and tmux, still covers any machine you can reach. The remote page below teaches each one, honestly and in order.

Where they run out is the same place every time: several machines, agents that are not Claude Code, sessions that must outlive the process that started them, a host you would rather not expose. That is the job spawnd was built for — one daemon on each host you own, dialing out, and any browser as the console, with sessions that live on the host and a phone that gets the permission prompt. The pages in the rack below teach the searched-for thing first; the product enters only where the honest route stops.