Claude Code Without an IDE: The Case for Pure CLI and Unix

There is a question that resurfaces whenever someone adopts Claude Code: do I integrate it into my editor or use it from the terminal? Most people answer with VS Code, Cursor, or similar. But in early April, the developer behind sindro.me published a detailed description of their setup in which the answer is clear: pure CLI, pure Unix, zero IDE. The thread reached Hacker News in early May, and although it had few points and no comments yet, the article circulated enough to deserve attention.

It is not a provocative stance on a whim. It is an engineering decision with concrete arguments.

What the setup proposes

The author works directly with Claude Code in the terminal, relying on classic Unix tools: `tmux` for managing sessions and panes, `vim` or `neovim` as an editor when you need to touch code manually, and shell scripts to automate parts of the workflow. No IDE extensions, no floating AI panels, no proprietary workspace configurations.

Integration with Claude Code happens through its standard command-line interface, leveraging the capabilities that Anthropic has been adding to the product: support for MCP servers, configurable hooks on lifecycle events (`PreToolUse`, `PostToolUse`, `Stop`...), and the ability to invoke subagents for delegated tasks. All without leaving the terminal.

The workflow described is roughly this: you open a `tmux` session, start Claude Code in one pane, work in another pane with your editor or with `grep`, `awk` and friends, and let Claude Code operate on the file system and execute commands autonomously while you supervise the output. The hooks allow you to trigger custom actions—tests, linters, notifications—in response to what the agent does.

Why it makes sense for certain profiles

The proposal is not for everyone, and the author themselves do not intend it to be. It makes sense in specific contexts:

• Remote servers and headless environments. When you work via SSH on a machine without a graphical interface, an IDE is not a real option. Claude Code in CLI is.
• Infrastructure and scripting projects. If your daily work is pipelines, configurations, and automation, the natural environment is already the terminal.
• Developers who prioritize explicit control. Without graphical abstraction layers, every action the agent takes is visible and interceptable. Claude Code's hooks fit well here: you can audit or stop any operation before or after it happens.
• Machines with limited resources. An IDE with AI extensions consumes memory. A terminal does not.

What you gain is predictability and composability: Claude Code behaves like any other Unix tool, whose output can be redirected, piped, or logged. What you lose is the integrated visual experience—graphical file navigation, inline diffs, visual autocompletion—that many developers consider essential.

The broader context

These kinds of configurations documented by the community are valuable precisely because Anthropic does not always detail extreme use cases in its official documentation. Claude Code was designed to be composable, and setups like syndro.me's demonstrate that this composability works in practice, not just on paper.

It also points to a real trend: as agents gain the ability to operate autonomously over complete systems, with access to files, shell commands, and external services via MCP, the IDE as the primary interface loses some of its centrality. The agent does not need the IDE; you needed it to read and write code. If the agent does a growing share of that reading and writing, the equation changes.

This does not mean IDEs will disappear—most teams will keep using them for a long time—but it does mean the CLI becomes a first-class option again for those who know how to use it.

---

EP Opinion: An unassuming post that documents something useful. The value is in the configuration details, not the thesis. If you regularly work remotely or in server environments, it is worth reading before installing another IDE extension.