Essential Configuration

You have Cursor installed and your project open. You type a prompt, the AI suggests running npm test to verify its changes, and then… it stops and waits for your approval. You click “Allow.” It suggests mkdir -p src/utils. It stops again. You click “Allow.” It wants to run tsc. You click “Allow.” Twenty clicks later, you wonder why anyone calls this productive. The fix takes sixty seconds: enable auto-run mode. Combined with Max mode for complex tasks and proper privacy configuration, these three settings transform Cursor from a chatbot-with-an-editor into a genuine development partner.

What You’ll Walk Away With

• Auto-run mode configured with sensible allow/deny lists so the AI executes safe commands without asking
• Max mode understood and ready to toggle for complex tasks requiring deep codebase analysis
• Privacy settings locked down for your environment — whether personal projects or enterprise compliance
• A clear mental model for which setting to change when the AI feels too slow, too limited, or too risky

Auto-Run Mode (Formerly “YOLO Mode”)

Auto-run is the single most impactful setting in Cursor. It lets the agent execute terminal commands — running tests, creating directories, building your project — without stopping to ask permission for each one.

Enable Auto-Run

1. Open Cursor Settings: Cmd+Shift+J (macOS) or Ctrl+Shift+J (Windows/Linux)

2. Navigate to the Agent section

3. Toggle Auto-run to enabled

4. Configure your auto-run prompt with allowed commands

Configuring Your Allow List

When you enable auto-run, Cursor gives you a text field to describe what commands are safe to run automatically. This is not a JSON config — it is natural language that the agent interprets.

Copy-paste auto-run prompt

Paste this into your auto-run configuration:

Any kind of tests are always allowed: vitest, jest, npm test, pytest, cargo test, go test.
Build commands are always safe: npm run build, tsc, vite build, cargo build.
File operations are fine: mkdir, touch, cp, mv (but not rm -rf).
Git read commands are safe: git status, git diff, git log, git branch.
Linting is always ok: eslint, prettier, biome, ruff, clippy.
Package installs are fine: npm install, yarn add, pip install.

What to Deny

Equally important is telling the agent what it should never run without asking:

Caution

Commands you should always require approval for:

• rm -rf — destructive file deletion
• git push — pushing to remote repositories
• git reset --hard — discarding uncommitted changes
• npm publish — publishing packages
• docker rm — removing containers
• Any command involving production databases or deployment

Auto-Run in Practice

Here is the workflow difference:

Without auto-run: You prompt the agent to “fix the TypeScript errors.” It finds the errors, proposes fixes, suggests running tsc to verify… and waits. You click Allow. It finds another error, fixes it, wants to run tsc again… and waits. Ten interruptions for one task.

With auto-run: You prompt “fix all TypeScript errors and verify the build passes.” The agent runs tsc, reads the errors, fixes them across files, runs tsc again, sees zero errors, and reports back. You review the diff once.

Copy-paste prompt for auto-run test-driven workflow

Open Agent mode (Cmd+I) and paste:

Write tests for the authentication module first, then implement the code
to make all tests pass. Run the tests after each change and keep iterating
until everything is green.

With auto-run enabled, the agent will create tests, run them, see failures, write implementation code, run tests again, and iterate — all without stopping to ask permission.

Max Mode

Max mode extends the context window beyond the default 200k tokens. For models like Gemini 3 Pro, it unlocks the full 1M token context. For Claude models, it extends to their maximum available window.

When to Enable Max Mode

Use Max Mode	Skip Max Mode
Refactoring that touches 50+ files	Single-file bug fixes
Architecture analysis of a large codebase	Writing a new utility function
Planning a feature that spans multiple modules	Documentation updates
Debugging an issue that involves multiple services	Routine coding tasks

How to Toggle Max Mode

Click the Max toggle next to the model picker in the agent panel. You can also toggle it per-conversation — it does not have to be a global setting.

Note

Max mode is slower and consumes more tokens (more expensive). Think of it like switching from a quick scan to a deep analysis. Use it when the default context is not enough, not as a default for everything.

Model-Specific Context Windows

Model	Default Context	Max Mode Context
Claude Opus 4.6	200k	1M
Claude Sonnet 4.5	200k	1M
Gemini 3 Pro	200k	1M
GPT-5.2	272k	—
Grok Code	256k	—

Max mode is most impactful with Claude and Gemini models where it significantly extends the available context.

Privacy Configuration

Whether you are working on a personal side project or enterprise code under NDA, understanding Cursor’s privacy settings is non-negotiable.

Enable Privacy Mode

1. Open Settings: Cmd+Shift+J / Ctrl+Shift+J

2. Navigate to General then Privacy

3. Enable Privacy Mode

4. Verify the status reads “Privacy Mode enabled”

When Privacy Mode is on:

• Your code is not stored on Cursor’s servers beyond the duration of a request
• Your code is not used for training any AI models
• Conversations are not persisted server-side

Note

Privacy Mode does not change functionality. You still get the same AI quality. It only changes data retention and training policies. There is no reason not to enable it.

Telemetry Settings

For maximum privacy, also review these in Settings:

• Usage analytics — disable if you prefer no telemetry
• Crash reporting — can be disabled, though it helps the Cursor team fix bugs

Enterprise Considerations

For teams on Cursor’s Business or Enterprise plans:

• Privacy Mode is enforced organization-wide by default
• Team admins can set mandatory rules via the Cursor dashboard
• SOC 2 compliance documentation is available on request
• No code leaves your machine in plaintext — all API communication is encrypted

Copy-paste Cursor rule for sensitive projects

Add this as an always-apply rule in .cursor/rules/privacy.mdc:

---
description: Security and privacy standards
alwaysApply: true
---

# Privacy Requirements

- Never include API keys, tokens, or secrets in generated code
- Use environment variables for all sensitive configuration
- Do not log personally identifiable information (PII)
- Exclude .env files, credentials, and private keys from any file operations

Quick Settings Checklist

Before moving on, verify these settings are configured:

Setting	Location	Recommended Value
Auto-run	Agent settings	Enabled with allow list
Privacy Mode	General > Privacy	Enabled
Max Mode	Agent panel toggle	Off by default, toggle per-task
Auto-save	File settings	On (essential for smooth AI workflows)
Format on Save	Editor settings	On (keeps generated code clean)

When This Breaks

Auto-run executes something unexpected: Cursor creates a checkpoint before every set of changes. Use the restore button in the chat panel to roll back. Then tighten your auto-run allow list.

Max mode is too slow: Switch back to normal mode for the current task. Max mode adds latency because the model processes more context. Only use it when you actually need the larger window.

Privacy Mode blocks a feature: Privacy Mode should not block any features. If you experience issues, check that your Cursor subscription is active and your network allows connections to Cursor’s API endpoints.

AI responses seem worse after changing settings: Clear the current chat and start fresh. Settings changes do not retroactively affect an ongoing conversation, so old context may conflict with new behavior.

What’s Next

Model Selection Learn when to use Claude Opus 4.6, Sonnet 4.5, Gemini 3 Pro, or GPT-5.2 -- and how to switch between them mid-task.