Vibe coding with Claude Code course Β· Module 15: Advanced Claude Code

Claude Code beyond the terminal: app, cloud, phone, IDE and Chrome

8 min read
In this lesson9

The terminal is a good gangway but you do not always stand on it. Sometimes you would rather look at the changes in a window than read them on a black screen. Sometimes the task will take an hour and you have to close the laptop. Sometimes you only have your phone with you. And sometimes a bug shows only in the browser and is easier to show than to describe. Claude Code has a separate deck for each of these situations.

One engine, many decks

The same agent you know from the terminal works everywhere: it reads files, edits them and runs commands just the same. All that changes is where the session runs and what you talk to it through. On your computer (terminal, app, editor) the settings, the project memory in CLAUDE.md and the MCP servers are shared, so you never configure anything twice. A cloud session works on a fresh clone of the repository, not on your disk: it knows the CLAUDE.md and .claude/settings.json from the repo, but not configuration that lives only on your machine, such as ~/.claude/CLAUDE.md or user settings. So whatever has to apply everywhere, keep in the repository.

Desktop app: starting without a terminal

The Claude desktop app has a Code tab, which is Claude Code with a graphical interface. You do not install Node.js or the CLI: you sign in, pick the Local environment, point at the project folder and write what needs doing. You review changes in the diff view, file by file, and you watch the running app in a browser panel in the same window. The app runs on macOS and Windows, and on Linux as a beta (as of November 2026), and like the terminal it needs a paid plan. Scripts with claude -p stay in the terminal, but for everyday work with the agent the app is perfectly enough.

Cloud session: work goes on while you close the laptop

A cloud session runs on an Anthropic virtual machine, not on your computer, so it keeps working after you close the laptop. For it to clone the code and push a branch, you connect your GitHub account once at claude.ai/code. After that the order is always the same: you push your commits, send the task, Claude clones the repository and works, and in the end it pushes a branch with the changes, which you review before creating a Pull Request. You can send a task from the browser (you pick the repository under the prompt box) or from the terminal with the --cloud flag:

claude --cloud "Fix the login test in tests/auth.test.ts"

This creates a new cloud session for the repository you are standing in, and you keep working locally. The cloud machine clones your branch from GitHub, not a local folder, so it will not see commits you have not pushed. That is why you run git push first and --cloud second. Mind the bill too: a cloud session uses your plan's limit just like local work, although you do not pay separately for the machine. You can pull a finished session back into the terminal with claude --teleport, and in Cursor you read its diff like any other.

Phone: spyglass and remote helm

The Claude app for iOS and Android also has a Code tab. The phone runs nothing itself: it is a spyglass and a helm for sessions running elsewhere. You start a cloud session from it, check progress and answer the agent's questions. The other option is Remote Control, which steers from your phone a session running on your computer. You turn it on in an open session with one command:

/remote-control

From then on you find the session on the list in the Code tab on your phone and at claude.ai/code. The code still runs on your machine, with your files, tools and MCP servers. Remote Control does not move work to the cloud: the computer has to stay on and the claude process has to keep running. When the laptop goes to sleep, the session waits and reconnects only when the computer is back online. If the laptop is going to be closed, choose a cloud session.

Editor: VS Code, Cursor and JetBrains

If you prefer to stay in the editor, you have two routes. The VS Code extension gives you a graphical Claude Code panel right in the editor: in Manual mode you see every proposed change as a side-by-side comparison and can accept or reject it. It installs in Cursor too, so you do not have to abandon the editor you already know. The rule of this course does not change: in Cursor (or any other editor) you read files and the diff, and Claude Code writes the code, whether you talk to it in the terminal or in the extension panel. The JetBrains plugin (IntelliJ, WebStorm, PyCharm and others) works differently: it runs claude in the IDE terminal and shows changes in the IDE diff viewer, so it needs the CLI installed separately.

Chrome: a test in a real browser

Unit tests will not tell you that a button hid under the footer. The Claude in Chrome extension lets the agent open your app in a real browser, click, fill in forms and read errors from the console. You start a session with the browser using the claude --chrome flag (the /chrome command shows the connection state) and then ask in plain words:

Request for Claude Code:

Open localhost:3000, submit the login form with a wrong password
and check that an error message appears. Look at the console too.

Claude opens a new tab in a visible window, so you see every move. It uses your signed-in sessions, so treat this as handing over the helm: send it only where it is allowed to click. On a login screen or a CAPTCHA it stops and asks you. When it reports the result, ask for the list of steps it took and the exact text of the console message, and repeat the key step yourself in the browser.

A screenshot instead of a paragraph

A sketch on paper or a broken page layout tells the agent more than a paragraph of description. In the terminal you paste an image from the clipboard with Ctrl+V (in iTerm2 Cmd+V, on Windows and in WSL Alt+V), and an [Image #1] marker appears in the prompt box. In the VS Code panel you also paste an image from the clipboard, and in the desktop app you add it with the + button or drag it onto the prompt box. Add what you expect next to the image:

Request for Claude Code:

[Image #1] This is how the header looks on a phone today: the menu overlaps the logo.
Fix the styles in src/components/Header.tsx so the menu collapses under a button.

The image is input here, just like a file pointed to with @. It does not replace the goal, though: without a sentence about what should change, the agent guesses.

Output styles for learning

By default Claude simply does the task. When you want to learn along the way, change the output style, a set of instructions that sets how the agent works for the whole session. Two built-in styles were made for learning: Explanatory adds short Insight blocks in the conversation explaining why the code looks the way it does, and Learning also leaves you small pieces to write yourself. You switch them with a command:

/output-style explanatory

The explanations go only into the conversation, not into your files as comments. In Learning style Claude marks a spot in a file with a TODO(human) comment, describes the task and waits for you to write your piece; it still does the routine part itself. A bare /output-style lists the available styles, and /output-style default restores the normal mode. For the length of this course I recommend Explanatory: Claude still writes the code, and you read the diff in Cursor together with the explanation of why the change looks the way it does. Turn on Learning when you want to practise writing small pieces yourself.

Which deck when

SituationDeck
Scripts, automation, the full feature setTerminal (CLI)
Starting without a terminal, reviewing changes in a windowDesktop app, Code tab
Working in an editor you already knowVS Code extension (also in Cursor) or the JetBrains plugin
A long task with the laptop closedCloud session at claude.ai/code
Viewing and steering from your phoneMobile app: cloud sessions or Remote Control
Checking the app in a browserClaude in Chrome

My advice: keep working with Claude Code as you have so far, in Cursor's terminal or in the extension panel, and read every diff in Cursor before you accept it. Send long tasks that need no watching to the cloud, and let Claude in Chrome check how the page looks. In the next lesson we string up the safety net: you will see how to stop and undo a failed move of the agent.

Remember: there are many decks, but the crew is the same, so choose the one from which the goal of the voyage is easiest to see.

Code for this lesson: cloud-and-remote.sh
#!/bin/bash
# Claude Code beyond the terminal: cloud, phone and browser
# Claude Code always writes the code, and you read the diff in Cursor.

# --- Cloud session: keeps working after you close the laptop ---
# The cloud clones your branch from GitHub, not a local folder,
# so push your commits first.
git push

# Send a task to a new cloud session for this repository
claude --cloud "Fix the login test in tests/auth.test.ts"

# Pull the finished session back into the terminal
claude --teleport

# --- Remote Control: the phone steers a session on your computer ---
# Type in an open claude session:
#   /remote-control
# The computer has to stay on and the claude process has to keep running.

# --- Claude in Chrome: a test in a real browser ---
claude --chrome
# The connection state is shown in the session by the command:
#   /chrome

# --- Output style for learning (in a session) ---
#   /output-style explanatory   -> Insight blocks with explanations in the conversation
#   /output-style               -> list of available styles
#   /output-style default       -> back to the normal mode

echo "Claude writes, Cursor shows, you check the course."

Spotted a mistake in this lesson?

Check yourself

Answer the questions from this lesson. Pick an answer to see right away whether it is correct.

  1. 1. You steer a session running on your laptop from your phone through Remote Control. What happens when you close the laptop?

  2. 2. You send a task with the claude --cloud command from a repository on GitHub. What will the cloud session not see?

These are 2 of 3 questions for this lesson. Solve the rest in the game.

Hands-on tasks in the game

  • Vertical ordering

    Put the steps of working with a cloud session sent from the terminal in order:

  • Click in order

    Click the elements in order to send a task from the terminal to a new cloud session:

  1. QuestionQuiz: Remote Control and a closed laptop
  2. QuestionQuiz: What a cloud session does not see
  3. Put in orderSorting: A cloud task step by step
  4. QuestionQuiz: Which deck for a browser test
  5. Click in orderSequence: A task to the cloud

Useful articles