The Claude Code desktop app: a guide for people who don't use a terminal

The Claude Code desktop app is the easiest way to use Claude Code without a terminal. It is the Code tab inside the Claude desktop app for macOS and Windows, with Linux in beta. You choose a folder on your computer, type what you want in plain English, and review every change as a visual before-and-after comparison. Anthropic's quickstart puts it this way: the app lets you "ask Claude to work on the code in a folder on your computer and review its changes without using a terminal."

It is a good fit for business owners, marketers and content editors who want to make or review website changes themselves, with a developer or agency in the loop. It runs the same engine as the terminal version and reads the same project settings, so a team can mix both. What it does not do is scripted or unattended batch work. For that, Anthropic points you to the command line.

Chat, Cowork and Code: which tab does what

The Claude desktop app has three tabs, and Anthropic describes them this way:

• Chat: "General conversation with no file access, similar to claude.ai."
• Cowork: "An autonomous background agent that works on tasks independently while you do other work."
• Code: "An interactive coding assistant with direct access to your local files."

For website files, you want Code. Chat can draft copy but can't open your site's folder. If you are still unsure whether you need Claude Code at all, our guide to whether non-developers can use Claude Code covers what is realistic.

Setting it up

1. Install the app. Download it from Anthropic's desktop quickstart, run the installer and sign in with your Claude account.
2. Open the Code tab. If it asks you to upgrade, Claude Code needs a paid plan. The free plan does not include it.
3. Choose where Claude runs. Pick Local to work on a folder on your computer. The other choices are Cloud, SSH (a remote server you manage) and, on Windows, WSL.
4. Select your project folder. Choose the folder that holds your site's files, ideally a git repository and a copy of the site, not the live server.
5. Pick a model and a permission mode from the controls next to the send button. Start with Plan or Manual, as explained below.
6. Send a small first task. Anthropic suggests starting "with a small project you know well." One of its sample prompts is Create a CLAUDE.md with instructions for this codebase.

The check that proves setup worked: Claude answers a question about your folder, such as "list the page templates in this project," with file names that actually exist. Our step-by-step guide to your first Claude Code session on a website goes further.

Permission modes in the desktop app

The mode selector decides how much Claude can do without asking. As of October 2026 it offers four modes. A fifth, bypass permissions, appears only after someone turns it on in settings (Pro and Max) or through organization policy (Team and Enterprise).

The app remembers the mode you pick for each folder, except Plan, which applies to the current session only. Anthropic's own tip is to start complex tasks in Plan, then switch to Accept edits or Manual to carry the plan out. Our guide to Claude Code permission modes explains each mode and the settings behind it.

Reviewing changes without reading code

This is where the desktop app earns its place. After Claude edits files, a counter such as +12 -1 appears. Click it to open the diff view, with a file list on the left and the changes on the right.

• Comment on a line. Click any line, type what you want changed, and submit all your comments at once with Cmd+Enter on macOS or Ctrl+Enter on Windows. Claude revises and shows a new diff.
• Ask for a review. Type /code-review in the prompt box and Claude reviews its own changes before you commit, grouping findings by file. That is a second opinion, not a replacement for yours.
• Attach context. You can attach screenshots, PDFs and mockups, or type @ and a file name to point Claude at a specific file. The terminal version doesn't take file attachments.

A diff shows what changed in the code, which is not the same as how the page looks. Check the rendered page too.

Previewing your site in the Browser pane

The Code tab can start your site's local development server and open it in a Browser pane. Anthropic describes Claude verifying its own changes there: "it takes screenshots, inspects the DOM, clicks elements, fills forms, and fixes issues it finds." It stores the server settings in .claude/launch.json, which you can edit if your site uses a custom start command.

Two things to know. First, the Browser pane "uses a clean browser profile, separate from your personal browser, with none of your saved logins or history." That is the right place to test a page. When you need Claude to act inside sites you are logged into, such as your CMS or analytics, that is the Claude in Chrome extension, covered in our guide to Claude Code and Chrome. Second, on external sites Claude asks before acting, with Allow once, Always allow or Deny, and Anthropic says it won't "purchase items, create accounts, or bypass CAPTCHAs without your input."

Claude checking its own work is useful, but you should still look at the changed pages yourself, on desktop and mobile widths, before anything ships.

Sessions, pull requests and connectors

• Parallel sessions. Each conversation is a session with its own history and folder. For git projects, a worktree option gives a session its own isolated copy of the files, so two tasks don't overwrite each other.
• Pull requests. After Claude opens a pull request, a status bar shows your automated checks. You can turn on auto-fix for failing checks, or auto-merge once checks pass. Both need GitHub's gh tool installed, and auto-merge also needs it turned on in your GitHub repository. Keep auto-merge off for a production site unless a person has already approved the change.
• Connectors. The + button next to the prompt box adds connectors to services such as GitHub, Slack, Linear and Notion. Anthropic notes that "Connectors are MCP servers with a graphical setup flow." Each one gives Claude access to a real account, so add only what the task needs.
• Plugins. The same + menu installs plugins without the terminal, for example a team's shared review setup.
• Cloud hand-off. Select Cloud to run a long task on Anthropic's machines that continues after you close the app. See our guide to Claude Code on the web.

Scheduled tasks in the desktop app

The app's Routines page creates two kinds of recurring jobs. A Local scheduled task runs on your computer with access to your files. A Cloud routine runs on Anthropic's infrastructure even when your computer is off. Routines are in research preview as of October 2026.

For local tasks, as of October 2026:

• Presets are Manual, Hourly, Daily, Weekdays and Weekly. You can also describe a schedule in plain words in any session.
• Each task has its own permission mode. In Manual mode, a run stalls until you approve a missing permission.
• Tasks "only run while the desktop app is running and your computer is awake." If the computer sleeps through a run, it is skipped. When the computer wakes, the app starts one catch-up run for the most recently missed time in the last seven days.
• By default a task runs against your working folder as it is, "including uncommitted changes." Turn on the worktree option so each run gets its own isolated copy.

A sensible first scheduled task is read-only: "every Monday, list pages in this repository missing a meta description and write the list to a report file." Let it report, not edit, until you trust it.

What the desktop app can't do

• No scripting. In Anthropic's words, "Desktop is interactive only." There is no equivalent of the terminal's scripted mode for batch jobs. That lives in the terminal version.
• Missing features. Agent teams and inline code suggestions aren't available, and the dontAsk permission mode, meant for locked-down scripts, is CLI-only.
• Not a deploy tool. The app edits files. Getting them live still goes through your git workflow and host, and a person should approve each deploy.
• Same usage limits. Desktop draws on the same plan as the rest of your Claude use. Even the gray suggested next prompt is generated by a background request that counts toward your limits; you can turn suggestions off in settings.

If you already use the terminal, you can run both on the same project. Typing /desktop in the terminal moves a session into the app on macOS and x64 Windows.

A safe working pattern for non-developers

1. Work on a copy of the site in git, never the live server. Our guide to WordPress staging sites shows one way to set up a copy.
2. Start each task in Plan mode and read the plan before approving it.
3. Switch to Manual for the edits, and review each diff.
4. Preview the changed pages in the Browser pane and in your own browser.
5. Commit with a clear message, and have a developer or agency review before deploy.
6. Keep passwords and API keys out of prompts and out of CLAUDE.md.

Conclusion

The desktop app is the right Claude Code surface for people who want to see and approve every change without learning a terminal. Its diff view, line comments and Browser pane make review practical for non-developers, while Plan and Manual modes keep a person in charge. It is not built for scripted batch jobs or unattended work, and it does not remove the need for a copy of the site, git and a reviewer before deploy. If you want your team set up with the desktop app, sensible permission settings and a review process, our Claude Code consulting covers that rollout. More guides are in our Claude Code Guide.

Frequently asked questions