⚙️ Access & Setup

Access & Setup: Claude has three doors into your daily work: the claude.ai web chat, the Claude Code CLI running in your terminal, and IDE extensions for VS Code or JetBrains — a

Claude has three doors into your daily work: the claude.ai web chat, the Claude Code CLI running in your terminal, and IDE extensions for VS Code or JetBrains — and the mechanism behind that analogy is exact: all three are just different transports carrying your prompt to the same underlying model, the way ChromeDriver, FirefoxDriver and EdgeDriver are different transports carrying the same WebDriver API calls to different browsers. Here is the question worth sitting with: if they all talk to "the same Claude", why would a tester ever leave the comfortable web chat for a terminal tool that needs installing and a PATH that might break? Because the web chat cannot read your repository, run your test suite, or edit a file on disk — it only sees what you paste. Java comparison: it is the difference between pasting a stack trace into a colleague's inbox (web chat) and having that colleague sit at your keyboard with full access to your IDE and terminal (a CLI agent) — both can help, but only one can actually run mvn test and watch it fail. The QA stake is concrete: a junior who only ever pastes snippets into the web chat caps their own ceiling; a tester who installs the CLI can eventually let Claude read a failing test, run it, and propose a fix in the same terminal session — the difference between asking for advice and delegating a task.

Three Doors, One Model

claude.ai is the web chat: free to start, with a paid Pro tier that raises usage limits and unlocks longer conversations — the natural starting point for Q&A learning and one-off test case or test data generation, exactly what you practiced in the Prompt Engineering tab. Claude Code is a command-line agent: install it once, run it inside any project folder, and it can read files, run your test suite, and edit code with your permission — this is the tool that turns Claude from an advisor into an active participant in your repository. IDE extensions (VS Code, JetBrains) embed the same agent inside your editor, showing suggested edits inline instead of in a separate terminal window — convenient once you are already comfortable with the underlying CLI workflow.

Reasoning: if the CLI is strictly more powerful, why not always use it? Because power implies risk — a CLI agent that can edit files and run shell commands needs explicit permission settings, and running it against a live production branch without reviewing its diffs is asking for trouble. The web chat's read-only nature is a feature when you only want a second opinion on a test plan, not a code change. Senior testers pick the tool by blast radius: web chat for advice, CLI for supervised repository work.

Installing Claude Code (CLI)

Windows (PowerShell)

macOS / Linux (Terminal)

Why Is sudo npm install -g Risky, and Why Is nvm a Safer Alternative?

sudo npm install -g ... installs…

sudo npm install -g ... installs the package with ROOT privileges — this FLIPS ownership of every file npm writes to root, NOT to your regular user account.

The NEXT (sudo-free) npm install -g…

The NEXT (sudo-free) npm install -g command can CONFLICT with those previously root-owned files and PRODUCE an EACCES error — sudo may look like a "quick fix" but it stores up a new PROBLEM.

nvm (Node Version Manager) instead…

nvm (Node Version Manager) instead installs Node under the user's OWN home directory — this makes npm install -g work WITHOUT needing sudo AT ALL.