Claude Code

One engine, several places to sit

Claude Code is Anthropic's agentic coding tool: it reads a codebase, edits files, runs commands, and works across your other tools. What separates it from a single-surface product is where it runs.

The terminal CLI, VS Code and JetBrains extensions, a desktop app, the web, and the Claude mobile app are all the same engine. Your repository's CLAUDE.md, your settings and your MCP servers work on every one of them, so moving between surfaces does not mean reconfiguring.

That shapes how teams adopt it. You are not choosing between a terminal agent and an IDE assistant, which is the decision Codex and Windsurf present. You are choosing where to sit today.

Install and start

curl -fsSL https://claude.ai/install.sh | bash   # macOS, Linux, WSL
cd your-project && claude

Windows has PowerShell and CMD installers, and there are Homebrew, WinGet, apt, dnf and apk routes. One detail worth knowing: native installs update themselves in the background, and Homebrew and WinGet installs do not, so those need brew upgrade or winget upgrade to pick up fixes.

Homebrew also offers two channels. claude-code tracks stable, about a week behind and skipping releases with major regressions; claude-code@latest ships immediately.

The customisation layer is the part that pays

Four mechanisms, and they are what turn it from a chat window into something that knows your project.

CLAUDE.md sits in your project root and is read at the start of every session. Coding standards, architecture decisions, preferred libraries, review checklists. If your repository already has an AGENTS.md for other agents, Claude Code can read that instead or alongside. It also builds auto memory as it works, carrying learnings across sessions without you writing them.

Skills package a repeatable workflow your team shares, invoked like /review-pr or /deploy-staging.

Hooks run shell commands before or after its actions. Formatting after every edit, lint before a commit. This is where you enforce something rather than ask for it.

MCP connects it to your own systems, which is the protocol MCP in one sitting covers.

Of the four, CLAUDE.md returns the most for the least effort, and hooks are the one people reach for last and should reach for earlier, because a hook is a guarantee where an instruction is a preference.

Composable, which is the useful part in CI

It follows the Unix philosophy, so it pipes:

tail -200 app.log | claude -p "Slack me if you see any anomalies"
git diff main --name-only | claude -p "review these changed files for security issues"

-p makes it non-interactive, which is what makes GitHub Actions and GitLab CI integrations possible. Automated PR review and issue triage run through the same binary you use locally.

Running several at once

Subagents let one lead agent split work and merge results, which is the pattern multi-agent context sharing describes, along with its token cost. Background agents run several full sessions you watch from one screen.

The Agent SDK exposes the same tools for building your own agent, with your own orchestration and permission rules. That is the route when you want Claude Code's capabilities inside your product rather than in your editor.

Where the work can live

Sessions move. claude --teleport pulls a web or mobile session into your terminal, /desktop continues a terminal session in the desktop app for visual diff review, and Remote Control keeps a local session reachable from a phone. Routines run on a schedule in the cloud; desktop scheduled tasks run on your machine with access to local files.

The practical version: start something long on the web, check it from your phone, finish it in your terminal.

Reading documentation for something moving this fast

Same caution Windsurf raises. Surfaces, scheduling and the agent features here have all changed shape within a year, and any article about them ages quickly, this one included.

Check code.claude.com/docs before relying on a specific flag, and check claude --version against what the docs describe. The stable concepts are the ones worth building habits on: a file of project instructions, hooks for enforcement, MCP for your own systems.

Further reading

• Anthropic, Claude Code overview: surfaces, installation, and the customisation mechanisms.
• Codex: the neighbouring terminal agent, and sandboxing as its operating decision.
• Claude Agent SDK: building on the same machinery.
• MCP in one sitting: the protocol behind the tool connections.