Are you starting to use Claude Code immediately after installation? Honestly, that's risky.
Claude Code is an AI that reads and writes files, executes commands, rewrites code, and can even create Pull Requests. In other words, it's not just a conversation partner; it's an entity that actually performs actions on your PC.
Starting without checking the settings is like telling a new hire, "There are no specific rules, so just do whatever you want." The more capable the staff, the more they will move forward based on their own judgment if there are no rules. AI is the same.
I personally opened all the setting screens for the Claude Code desktop app and re-verified them one by one. There is a clear distinction between settings you should definitely check and settings you don't need to touch from the start.
By reading this article, you will know where to look in the Claude Code desktop app settings, which settings to change, and which ones to ignore.
The Claude Code desktop app settings menu has many items: General, Notifications, Personalization, Apps, Schedule, Billing, Data Control, Storage, Security, Parental Controls, and Account. Below those are Privacy, Usage, Features, Connectors, Claude Code, Cowork, and Claude in Chrome. Further down, in the "Desktop App" section, there are General, Extensions, and Developer.
Honestly, it's a lot. But there are only four places you need to look at first.
- Desktop App "General"
- "Privacy"
- "Claude Code"
- Permission Mode and CLAUDE.md at the start of a session
You don't need to touch the rest initially.
1. Desktop App "General": Building a Foundation for Safe Use
Scroll down the left menu to the "Desktop App" section and open "General." This is the foundation for using Claude Code safely.
Run at Startup
This determines if Claude Code launches automatically when you log in. Keep it off for now. You can turn it on once you start using it daily.
Quick Access Shortcut
A shortcut to send messages to Claude from anywhere. The default is "Double-tap Option." Check if it conflicts with other apps.
Voice Shortcut
A shortcut to talk to Claude. Default is "None." Voice input is very useful for giving detailed instructions like:
- "Check this content, but don't make any changes yet."
- "Always show me the diff before changing anything."
- "Just find the cause of this error. No fix needed yet." Claude Code can go too far if instructions are vague. Being able to give frequent instructions via voice is safer.
Menu Bar
Whether to show Claude in the menu bar. Leaving it on is fine.
Prevent Computer from Sleeping
This is surprisingly important. If your PC sleeps while Claude is performing a long task, the work stops. If you're letting it handle time-consuming tasks like multi-file reviews or code rewriting, keep this on. It's on by default.
Browser Use
From here on, these are permission-related settings. Be careful. Turning on "Allow all browser actions" lets Claude browse and operate any website in Chrome without confirmation. The settings screen warns this could put data at risk. I recommend keeping it off initially. If you turn it on, close unnecessary tabs and be extra careful with pages showing emails, payments, or customer info.
Computer Use (Beta)
This is critical. When on, Claude can take screenshots and operate your keyboard and mouse in authorized apps. It's off by default. Keep it off for now. If you turn it on, macOS requires "Accessibility" and "Screen Recording" permissions.
Denied Apps
Register apps you don't want Claude to access. If you use Computer Use, add mailers, password managers, and internal confidential tools here.
2. Privacy: Checking Data Handling
Location Metadata
Allows Claude to use rough location data to improve the experience. This is optional.
Help Improve Claude
Allows your chat and coding data to be used for training Anthropic's models. If you handle client code or confidential projects, it's safer to turn this off. It's off by default, but double-check.
3. Claude Code: Deciding Session and PR Behavior
Session State Classification
Allows Claude to automatically classify sessions as blocked, pending review, or completed. Keep it on to track multiple parallel sessions.
Allow Bypass Permission Mode
Warning: This bypasses all permission checks. It's high risk and can lead to data loss or system damage. Keep it off. Don't use it outside of sandboxes or VMs.
Worktree Location
The desktop app automatically isolates a working copy in a Git worktree for each new session. This prevents changes in Session A from affecting Session B. The default is "Inside project (.claude/w...)".
Crucially, remember .worktreeinclude. Files ignored by git (like .env) aren't copied to worktrees automatically. If you get errors about missing environment variables, create a .worktreeinclude file in the project root and specify the files to copy.
Pull Request Automation
There are settings to automatically create PRs and auto-fix CI failures. Keep these off initially. It's safer to check the PR content yourself first.
Auto-archive after PR Merge/Close
I recommend turning this on. it prevents finished sessions from cluttering your sidebar.
4. Permission Mode and CLAUDE.md: Check Every Session
Permission Mode
Select this next to the send button at the start of a session. It determines how much Claude can do on its own.
- Ask permissions: Confirms every file edit and command. Best for beginners to understand how Claude works.
- Auto accept edits: Automatically accepts file edits and basic operations. Speeds up work once you trust Claude's judgment.
- Plan mode: Reads files and runs commands to investigate but doesn't edit code. Use this to get a plan first before switching to Auto accept edits.
- Auto: Almost fully automatic. Only for Max/Team/Enterprise/API plans. Don't start with this; Claude might misinterpret instructions and rewrite too much.
CLAUDE.md
If Permission Mode is "how much it can do," CLAUDE.md is "how you want it done." Place this file in your project root. Claude reads it at the start of every session. Use it for:
- Basic project commands (build, test, dev).
- Rules (e.g., "Always show diff before changing," "Don't change file structure").
- Correction rules (add a line whenever Claude makes a mistake to prevent it from happening again).
Keep it under 100 lines. Use /init in a session to auto-generate one.
Summary
Claude Code isn't just a chat AI; it's an agent operating on your PC. Spending 10 minutes on settings makes the difference between being controlled by the AI and being the one in control. Don't just say "Fix this"; start with "Check this and propose a plan." Safety first.
