Cursor Setup: How to Set Up Cursor AI

Head to cursor.com and click the download button. Cursor detects your OS and hands you the right installer: a .dmg for macOS (Apple Silicon and Intel), a .exe for Windows.

On Linux you have a choice. Cursor ships apt packages for Debian and Ubuntu and dnf packages for RHEL and Fedora, and the official docs recommend those over the portable AppImage because they add desktop icons, automatic updates, and the CLI tool.

Mac: Open the .dmg and drag Cursor to Applications. Requires macOS 12 (Monterey) or later.

Windows: Run the .exe installer. Defaults are fine. Requires Windows 10 or later.

Linux (Debian/Ubuntu): install via apt following the repo instructions on the download page. For the portable route:

chmod +x Cursor-*.AppImage
./Cursor-*.AppImage

On first launch Cursor walks you through a short setup: keyboard layout, theme, and the VS Code import. Say yes to the import if you have VS Code. Also let it install the cursor shell command so you can open any folder from the terminal.

Cursor needs an account for the AI features. Sign up with Google, GitHub, or email. Sign-in bounces through your browser and back to the app.

What the plans actually get you, per the live pricing data we track (verified August 20, 2026):

• Hobby (free): Tab completions and a limited number of Agent requests. No credit card. Enough to evaluate Cursor on a real project.
• Pro ($20/mo): extended Agent limits and access to frontier models. The plan most individual developers land on.
• Pro+ and Ultra: higher multiples of the Pro limits for heavy agent users. Cursor does not list their prices on the public pricing page.
• Teams ($40/user/mo): centralized billing, shared rules, SSO.

Full breakdown on our Cursor pricing page, which pulls from the same tracker.

Press Cmd+I (Mac) or Ctrl+I (Windows/Linux) to open the Agent sidepanel. This is the main AI surface in current Cursor. The old "Composer" and separate chat panel from earlier versions have been folded into it, so if a tutorial tells you Cmd+L opens something different, it is out of date. Both shortcuts now toggle the same panel.

The Agent has modes, and switching them is the skill worth learning first:

• Agent mode: reads your codebase, plans, edits multiple files, runs terminal commands, and shows you every change as a diff.
• Plan mode: press Shift+Tab in the input to rotate modes. Plan mode makes Cursor lay out its approach before writing code. Use it for anything bigger than a one-file change.
• Inline edit: select code and press Cmd+K for a quick targeted change without opening the panel.

And always: Tab accepts autocomplete suggestions as you type. Cursor Tab predicts whole edits, not just the next token.

The model picker sits in the Agent input, and Cmd+/ cycles between models without touching the mouse. Cursor routes to multiple providers:

• Claude (Anthropic): best for complex reasoning, refactoring, and understanding large codebases. The default recommendation for real development work.
• GPT (OpenAI): fast and capable. Good for quick edits and general-purpose tasks.
• Gemini (Google): strong long-context handling when you need many files in scope at once.

Open a folder with File > Open Folder, or from the terminal:

mkdir my-first-cursor-project
cd my-first-cursor-project
cursor .

Cursor indexes the project in the background so the Agent can search it. Small projects index in seconds. On big monorepos, add a .cursorignore file to keep generated folders out of the index.

Then open the Agent (Cmd+I) and try a real prompt. In an empty project:

Create a simple Express.js server with a health check
endpoint at /api/health that returns { status: "ok" }.
Include a package.json with the right dependencies.

In an existing project, start by asking instead of editing:

@codebase How is authentication handled in this project?
Which files are involved?

Cursor shows every change as a diff. Review, adjust if needed, then accept.

Rules tell Cursor about your stack and conventions so every prompt starts from the right assumptions. Current Cursor gives you two ways to do it:

• AGENTS.md: a plain markdown file at your repo root. The simple option, and the one to start with. Nested AGENTS.md files in subdirectories work too.
• .cursor/rules: a directory of .mdc files with frontmatter that controls when each rule applies (always, per file glob, when the Agent judges it relevant, or only when @-mentioned). Use this once you want different rules for different parts of the codebase.

A practical starter AGENTS.md:

# AGENTS.md

## Tech Stack
- Framework: Next.js 15 (App Router)
- Styling: Tailwind CSS
- Database: Supabase (Postgres)
- Language: TypeScript (strict mode)

## Conventions
- Functional components with hooks
- Prefer server components; add 'use client' only when needed
- Absolute imports from the @/ prefix
- Wrap async operations in try/catch
- Files in kebab-case, components in PascalCase

## Do Not
- Use CSS-in-JS libraries
- Create files over 300 lines; split into smaller modules
- Use default exports for non-page components

Now put the Agent to work on a feature that spans files. Open it with Cmd+I, switch to Plan mode with Shift+Tab if the change is non-trivial, and describe the outcome:

Add a user profile page at /profile that:
- Shows the user's name, email, and avatar
- Has an "Edit Profile" button that opens an inline form
- Saves changes via a PATCH /api/users/me endpoint
- Shows a success toast after saving
- Handles loading and error states

The Agent plans the change, creates or edits the files, and presents a complete diff. Review each file before accepting. Then ask it to verify its own work: run the type checker, the tests, the build. Cursor can run those commands itself and fix what breaks.