How to use Cursor AI: a first-hour tutorial

How do you install Cursor and set it up?

Cursor's quickstart lists the installers: a .dmg for macOS 12 (Monterey) and later, with Apple Silicon and Intel support; an .exe for Windows 10 and later; and Debian, Ubuntu, RHEL, Fedora and AppImage packages for Linux. The apt and yum packages are preferred over AppImage because they provide desktop icons, automatic updates and CLI tools. Everything here was checked against Cursor's own docs on October 30, 2026.

1. Download Cursor from cursor.com, install it and sign in. There is a free plan, so you can follow this tutorial without paying. See what is Cursor AI for the plans.
2. If you use VS Code, import your setup: open Cursor Settings with Cmd+Shift+J (Ctrl+Shift+J on Windows and Linux), go to General, then Account, and click Import under VS Code Import. It brings over extensions, themes, settings and keybindings. Cursor uses the Open VSX extension registry, so a few VS Code extensions may be missing.
3. Turn on Privacy Mode in the same settings window, under General, if you do not want your code used for training. For teams it is on by default.
4. Pick a folder to work in. Use an existing project or an empty folder for a new one. Start with something small and low risk.

How do you ask Cursor to do something?

Open Agent with Cmd+I or Ctrl+I, type a request in plain language and press Return. Agent explores your codebase and starts making changes, and you watch the edits appear in the diff view as they happen. You can press the Stop button at any time to redirect it. Cursor's own first-task advice is to begin with orientation, not a build.

Explain this codebase. Point me to the main entry points, key modules, and anything I should read before making changes.

Suggest three small, safe improvements in this codebase. Explain the tradeoffs and wait for me to choose one.

Good first tasks are low risk: improving some copywriting or fixing a small UI issue. Pick one of the three, tell the agent to do it, then review. If the agent heads the wrong way, stop it, hover over an earlier message and click Restore Checkpoint to roll back every file it changed after that point. Restoring reverts files only, not the conversation.

How do you review what Cursor changed?

The agent applies its edits as it works, and you review them in the diff view and reject anything you do not want. Treat this as the main skill of using Cursor: the agent writes quickly, and you decide what stays.

• Read every diff before you move on. Look for changes you did not ask for, deleted code and new dependencies.
• Ask it to run your checks. Cursor's quickstart says to ask the agent to run the checks your project already uses: tests, the type checker, linting or a local build.
• Use Agent Review for a second pass. Type /agent-review in the agent input, or open the Source Control tab to review all local changes against your main branch. It has a Quick depth for small diffs and a Deep depth for complex or security-sensitive code.
• Commit in small steps. Each working step is a point you can return to if the next one goes wrong.

How do Tab, @ mentions and selected code work?

Three habits make Cursor noticeably more useful once the basics work.

Type @ in the chat input to attach files, folders, terminal output, earlier chats, git diffs or the built-in browser, such as @auth.ts or @Branch (Diff with Main). Use mentions when you know which files matter. If you are not sure, skip them, because the agent finds relevant files through its own search. A context ring next to the prompt shows how full the model's context window is, and Cursor compresses older parts of a long chat into a summary when it fills. If a conversation gets long and muddled, start a new chat with a short summary instead of pushing on.

Tab works as you type, with no prompt needed. After accepting a suggestion, press Tab again and it jumps to the next likely edit, including in other files. You can snooze or disable it from the Tab indicator in the bottom-right corner.

When should you use Plan Mode?

Use it when a task spans several files, needs research or has more than one valid approach. Press Shift+Tab in the agent input. Cursor researches your codebase, asks clarifying questions, writes an implementation plan and waits for your approval before it builds. The Plan Mode docs add that plans are saved in your home directory by default and that you can click Save to workspace to keep them with the project.

One piece of advice from those docs is worth repeating. When the agent builds something that is not what you wanted, do not patch it through follow-up prompts. Revert the changes, make the plan more specific and run it again. That is usually faster and gives cleaner results.

Plan a feature: add a notes field to each booking in this app.

User story: a gym owner opens a booking and can add a free-text note that is saved with it.
Constraints: no new dependencies, keep the existing styling, add or update a test.
Out of scope: search, attachments, multiple users.

Ask me any questions you need, then write the plan. Do not change any code until I approve.

If you are starting from an idea and not a codebase, copy a ready build prompt from the ideas board and paste it into an empty folder in Plan Mode. The guide to writing an app build prompt explains what such a prompt needs, and the Cursor prompts guide collects more.

How do you make Cursor follow your conventions, and use it from the terminal?

When you see the agent make the same mistake twice, write a rule. Type /create-rule in Agent and describe what you want, and Cursor writes the file into .cursor/rules. Rules apply to Agent, not to Tab. The Cursor rules guide has the format and copyable examples, and Cursor reads AGENTS.md too.

The same agent runs in a terminal. Per the CLI docs, install it on macOS, Linux or WSL with the command below, then run agent to start an interactive session.

# install (macOS, Linux, WSL)
curl https://cursor.com/install -fsS | bash

# interactive session
agent

# start with a prompt
agent "refactor the auth module to use JWT tokens"

# one-off, non-interactive run (scripts, CI)
agent -p "review these changes for security issues" --output-format text

# modes: Agent is default; Plan with --plan; read-only Ask with --mode=ask

Inside a CLI conversation, start a message with & to hand it to a cloud agent, which keeps running while you are away. You can pick up cloud agent tasks on web or mobile at cursor.com/agents. The Cursor Agent guide covers running agents in parallel.

What does a good first hour with Cursor look like?

Once you can do that loop, you are using Cursor the way its docs intend. If you are still choosing between tools, Cursor vs Windsurf and Claude Code vs Cursor vs Lovable put it next to the alternatives.