3. Setting Up Your Environment

Claude Code can be used with various interfaces, the CLI running on a terminal being the initial and most common use case of it.

Claude Code runs in all the terminals, but it integrates well with some specific ones. The integration with the terminal affects your experience for working with the agent. How line-breaks are handled, how notifications are hooked up depends on which terminal you use.

I personally prefer using iTerm2 as it has been my terminal of choice for more than a decade. And, that's a first-class terminal to use with Claude Code. Other notable first-class options are: WezTerm, Ghostty, and Kitty.

You could also use editor terminals on VS Code, WebStorm, Alacritty, Zed, and Warp. That's up to you. Just execute /terminal-setup in your editor terminal if you do, this is not required for other terminals.

Claude fires a notifications event to your system (via Notification hook) when it finishes working or awaits your input (via Stop hook).

Ghostty and Kitty terminals support desktop notifications by default. iTerm 2 requires a small setup.

Head over to Claude Code setup docs to follow the installation guide.

Once installed, verify the installation in your terminal:

If you see a version number, you are ready to authenticate.

Once you launch claude for the first time in your terminal, Claude Code opens a browser window for you to log in. There are various account types you can authenticate with your subscription.

Authentication comes in three flavors:

Your choice here affects both cost structure and daily workflow. With an API key, you pay per token. With a Claude Pro, or Max subscription, you get bundled usage. Pro works well for light exploration and learning. Max is where professional daily use becomes cost effective, because the bundled token allocation covers most workloads without the pay-per-token anxiety.

Here is how to think about the decision:

You'll learn how to make the best out of your usage limits in Chapter 9 with Context and Cost Engineering practice.

After authenticating, verify everything works by asking Claude a simple question:

If you get a coherent response, your installation and authentication are working. Your CLI is running.

To run diagnostics and verify that you have no misconfigurations or any anti-patterns related to your installation, run /doctor command in Claude Code.

Before you start working, you need to know how to steer the CLI efficiently.

Claude Code CLI can be either invoked with one-off execution via the -p flag with a single prompt, or invoked for the interactive mode.

See the complete CLI reference below to get familiar with the flags and capabilities.

Claude Code is a conversation-driven tool, especially in the beginning where you still need to build your system around it and course correct its behavior. Knowing when and how to interrupt, redirect, and reset is as important as knowing what to ask.

The single most important habit I can teach you at this stage: interrupt early and often when things go off track.

These are the shortcuts you will often use in your sessions:

And, these are some of the terminal shortcuts you'd find useful to edit your prompt:

These commands control your session state:

Watch for drift signals during a session. When you first start working with AI coding agents, sessions often go off-track.

When you see any drift, press Esc immediately. Do not wait to see if Claude self-corrects. It costs less to redirect early than to untangle a wrong-direction implementation that touched 15 files.

After pressing Esc, give Claude a clear correction:

Claude is already aware of your interruption of its work, and has the full context before the interruption.

Claude keeps auto-memory of its learnings across sessions. However, to build reliable agentic systems, you should save any correction after the agent's mistake to the project memory, in a modular and systematic way for larger projects. You will explore this approach in Chapter 5: Guardrails for Quality.

When Claude is generating a response, you can type your next prompt immediately. It queues up and sends automatically when the current response completes. This is useful when you spot something mid-generation that you want to follow up on.

You can also edit a queued message before it sends. Just hit Up Arrow key to edit your queued messages, or /rewind (Esc + Esc keys) to rewind to a checkpoint before your queued messages.

Three scenarios, three responses:

Open a Claude session now and practice.

You can navigate and interrupt. Now set up the environment where you will review Claude's work.

Claude Code was originally built as a terminal agent but it's not limited to a terminal. The agent is available across many channels and interfaces. Depending whether you work in a terminal or an editor, you will get a similar but different "Claude Code" experience.

You have already installed the CLI. To discover the differences of CLI with other channels, and to integrate Claude to where you work install the apps and extensions that fits your workflow:

Claude integrates with the code editors to share the editor context. It knows about your current file and text selection, and proposes changes directly inside your editor window with the diff view.

These are the benefits of integrating your code editor with Claude Code:

Each setup has different strengths. Pick the one that fits where you already work.

When you select a file or highlight a line range in your IDE and then invoke Claude, it automatically picks up that context via /ide. This means you can point Claude at a specific function and say "refactor this" without typing the file path.

You get the diff view for the approval of each edit Claude makes in your integrated editor by default.

You could switch between the permission modes and plan mode using Shift + Tab key and set Accepts Edits On mode which will accept all the edits by default and would not prompt you for reviewing each edit.

When you auto-accept edits of Claude, review your local git changes with /diff within Claude's interactive CLI or git diff. Alternatively, use your editor's built-in diff view to review them before committing. In VS Code, the Source Control panel shows all modified files. In Zed, the git gutter highlights changes inline.

The Claude Code browser extension gives Claude eyes on your web application runtime, and it is a fantastic milestone towards the feedback loop you will create. Install it from the Chrome Web Store if you haven't yet.

With the browser extension, Claude can navigate pages to test your app, check layouts, fill forms and read console errors right in your active Chrome/Edge browser instance.

claude-in-chrome is a browser MCP server, and once enabled, browser actions become part of your coding workflow. I use a browser MCP constantly when building web apps:

This is the first step in what Chapters 5-6 will build into a complete verification system. You gave Claude eyes to see the runtime of your project.

Claude Code on the web lets you kick off sessions directly from the Claude app without any local setup. It runs on Anthropic managed cloud infrastructure, so there is nothing to install or configure. It is the right choice when you want to:

Claude Code is also available on the Claude app for iOS and Android, so you can kick off tasks running on Claude cloud infra and monitor progress from your phone on the go.

The /teleport command moves an active terminal session to the browser, letting you monitor progress from any device.

You can alternatively create a new web session on claude.ai from your terminal, and check progress from your phone.

Remote Control takes this further. Instead of just watching, you can leave your session behind running on your machine, and send messages and interact with your local Claude Code session from any browser on another machine or from the Claude mobile apps. Your terminal keeps running locally with your filesystem, MCP servers, tools, skills and project configuration. The browser or phone is just a window into that session.

Start a remote controllable session from your project directory:

Or, if you are already in a session, run /remote-control (or /rc) command to promote it. Claude prints a session URL, and the remote control session becomes visible on Claude Web and Claude Mobile apps. Open the URL on any device, find the session by name in claude.ai/code, or navigate to Code on your Claude mobile app.

The conversation stays in sync across every connected device. You can send a message from your terminal, continue from a browser tab, and check results on your phone. If your laptop sleeps or loses Wi-Fi, the session reconnects automatically when the machine comes back online.

I use this constantly when kicking off long running tasks. I start the task at my desk, walk away, and monitor the progress from my phone while taking a break from my desk.

You have reached the first milestone towards building your feedback loop with the Claude's browser MCP. But right now, every tool call prompts you for permission, breaking your flow and slowing down the agent.

Chapter 4 brings everything together: you will generate a project idea with your first skill, explore the .claude directory as your agent's operating system config, configure permissions so Claude works without interruption, and set meaningful defaults with claude config.