How to Use Cursor AI: Beginner’s Guide to Cursor 3, Agent, Rules and Pricing (2026)

Cursor is the AI code editor most developers try first, and in the last few months it changed a lot. It moved to Cursor 3 with a separate Agents Window, SpaceX bought it, its plans were merged into “The X Subscription” with Grok, and OpenAI models are due to leave on November 12, 2026.

This beginner-friendly guide shows you how to use Cursor AI step by step in November 2026: install it, pick the right plan, use Tab, inline edit and the Agent, write rules so it follows your project, add MCP tools, run cloud agents, and fix the problems people hit most.

What Cursor Is (and What Changed in 2026)

Cursor is a code editor built on VS Code with AI everywhere: it predicts your next edit, rewrites code you select and runs an agent that can read your project, edit many files, run terminal commands and fix its own errors.

Because it is a VS Code fork, your extensions, themes and keyboard shortcuts carry over. If you know VS Code, you already know 80% of Cursor.

The big changes this year, in order:

Not sure if Cursor is the right tool at all? Our Cursor vs Windsurf vs Claude Code vs Cline comparison covers the alternatives.

Cursor Plans and Prices in November 2026

Since October 12, 2026, new users sign up to The X Subscription, the same plan xAI sells for Grok. It bundles Grok, Cursor and X Premium, and all usage comes from one shared pool that refreshes weekly.

Two details catch people out:

• Third-party models cost extra. From Premium up you can use Claude and Gemini, but they are billed separately at API list prices, not from the shared pool.
• Old plans still exist. If you already pay for Cursor Pro ($20), Pro+ ($60), Ultra or Teams, you can keep it. A switch option appears gradually in billing settings with a prorated refund, and switching is permanent.

Prices change fast here. Check cursor.com/pricing before you pay; the figures above are from the October 12 launch of The X Subscription.

Step 1: Install and Set Up Cursor

1. Download Cursor from cursor.com for Windows, macOS or Linux (AppImage or .deb/.rpm) and install it like any app.
2. Sign in or create an account. New accounts start on Free.
3. Import from VS Code when asked. One click brings your extensions, themes, settings and keybindings.
4. Choose privacy settings. In Cursor Settings > General, turn on Privacy Mode if you do not want your code stored or used for training. Use it for work code.
5. Open a project folder: File > Open Folder. Cursor indexes the codebase in the background so the agent can search it. Watch progress under Settings > Indexing & Docs.
6. Install the command-line launcher (Command Palette > “Install ‘cursor’ command”) so you can open folders with cursor . from a terminal.

Step 2: Use Tab Autocomplete

Tab is Cursor’s autocomplete, and it is smarter than normal suggestions. It predicts multi-line edits and your next edit location, not just the rest of the line.

• Accept a suggestion with Tab. Reject it with Esc or just keep typing.
• Accept word by word with Ctrl/Cmd+Right Arrow.
• Jump to the next suggested edit by pressing Tab again. Rename a variable in one place and Tab walks you through the other places.

Tab completions do not use your agent pool on paid plans, so use them freely. Turn Tab off for certain file types (for example Markdown) from the status bar if it gets in the way.

Step 3: Edit Code With Ctrl/Cmd+K

Inline edit is the fastest way to change a small piece of code without opening a chat.

1. Select the code you want to change (or place the cursor on an empty line to generate new code).
2. Press Ctrl+K (Windows/Linux) or Cmd+K (Mac).
3. Type the change, for example “add input validation and return a 400 error if email is missing”.
4. Review the diff and accept or reject it.

In the terminal, Ctrl/Cmd+K turns plain English into a command, for example “find all files over 100 MB in this folder”.

Step 4: Build Features With the Agent

The Agent is where Cursor does real work. It reads your project, edits several files, runs commands, reads the errors and tries again.

1. Open the Agent with Ctrl/Cmd+I (Ctrl/Cmd+L also opens the chat panel and adds any selected code).
2. Pick a mode from the menu under the chat box:
   • Agent: full access to edit files and run commands. The default for building.
   • Plan: researches your code and writes a step-by-step plan you can edit before anything changes. Also available with /plan.
   • Ask: read-only questions about the code. Nothing is changed.
3. Describe the task clearly. Say what, where and how you will check it:

   Add a "forgot password" flow to the Next.js app.
   - API route in app/api/auth/reset/route.ts
   - Email via the existing lib/mailer.ts
   - Token expires after 30 minutes
   Run npm test when done and fix any failures.

4. Approve commands. By default the agent asks before running terminal commands. You can allow safe ones (like npm test) to run automatically in Settings > Agents.
5. Review the changes. Every edited file shows a diff. Accept or reject per file or per change, then commit.
6. Use checkpoints. If the agent goes wrong, click Restore checkpoint on an earlier message to roll the files back.

Which Model to Pick in Cursor

The model picker sits under the chat box. Auto lets Cursor choose. You can also pick one yourself:

After November 12: you can add your own OpenAI API key in Settings > Models to keep GPT in Chat and Agent. Tom’s Guide notes a key does not cover Tab, Auto, cloud agents, the CLI or the SDK.

Cursor’s co-founder said OpenAI was about 5% of Cursor’s model traffic, so most people will not notice.

Step 5: Teach Cursor Your Project With Rules

Rules are instructions the agent reads every time, so you stop repeating yourself. This is the single biggest quality boost for most people.

Option A: AGENTS.md in the project root. A plain Markdown file many coding agents understand. Cursor also reads nested AGENTS.md files in subfolders. Our AGENTS.md guide has full templates.

Option B: Project rules in .cursor/rules/, one file per topic, which can apply always, to matching files only, or when the agent thinks they are relevant. Create one from Settings > Rules or by asking the agent.

A good starter rules file:

# Project rules
- Stack: Next.js 15, TypeScript strict, Tailwind, Prisma + Postgres
- Package manager: pnpm (never npm or yarn)
- Run "pnpm lint && pnpm test" before saying a task is done
- Use server components by default; client components only when needed
- No new dependencies without asking
- Keep functions under 50 lines; write tests for new logic in __tests__/

Keep rules short and specific. Personal preferences that apply to every project (like “explain changes briefly”) go in User Rules under Settings > Rules.

Give the Agent the Right Context

The agent searches your code on its own, but pointing it at the right things saves time and usage. Type @ in the chat to add:

• @Files and @Folders: the exact code you mean.
• @Docs: library documentation you added under Settings > Indexing & Docs.
• @Web: a live web search for new APIs or errors.
• @Past chats: reuse decisions from an earlier conversation.
• Images: paste a screenshot of a bug or a design mock-up.

Start a new chat for each task. Long chats fill the context and the agent gets slower and less accurate. Cursor summarises long chats automatically, but a fresh chat with a clear brief is better.

The Agents Window, Worktrees and Cloud Agents

Cursor 3 added the Agents Window, a separate view that shows every agent you have running, including ones started from Slack, GitHub, Linear, the web or your phone. Open it from the Command Palette (Ctrl/Cmd+Shift+P > “Agents Window”).

• Projects group related tasks, with list and board views (added in version 3.27).
• Worktrees: type /worktree in an agent chat to give that agent its own Git worktree, so several agents can work without overwriting each other. /apply-worktree brings the result back.
• Best of N: /best-of-n runs the same task on several models, each in its own worktree, so you can keep the best result. Each model is billed separately.
• Cloud agents run on Cursor’s Ubuntu VMs, so you can close your laptop. They need a paid plan with cloud agents (Premium and up on the new plans). Define the setup in .cursor/environment.json and put secrets in the Secrets tab, never in the repo.

Prefer the terminal? The Cursor CLI (agent) runs the same agent in any shell and in CI.

Add Tools With MCP

MCP (Model Context Protocol) servers give the agent new abilities, such as reading your database, opening GitHub issues, browsing a page or checking error logs.

1. Open Settings > Tools & MCP and click Add, or install one from the marketplace.
2. For a manual setup, edit ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (this project):

   {
     "mcpServers": {
       "github": {
         "command": "npx",
         "args": ["-y", "@modelcontextprotocol/server-github"],
         "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token" }
       }
     }
   }

3. Restart the agent chat. The new tools appear, and the agent asks before using them.

Only add MCP servers you trust, because they run on your machine with your tokens. Our list of the best MCP servers is a good place to start, and our extension safety check explains how to vet what you install.

Bugbot, Live Apps and Voice

• Bugbot reviews your GitHub pull requests and leaves comments on likely bugs. It is usage-based, roughly $1.00 to $1.50 per review, and comes with Premium and up.
• Live apps (November 5) build a hosted app with database, login and a shareable link from one prompt. See our Cursor live apps guide.
• Voice chat with agents (November 6) lets you talk to the agent, check progress and hear updates while it works.

A Simple Daily Workflow

1. Create a branch for the task.
2. Plan: open the Agent in Plan mode, describe the feature and edit the plan until it is right.
3. Build: switch to Agent mode and say “implement the plan, run tests and fix failures”.
4. Review: read every diff. Use Ctrl/Cmd+K for small fixes and Tab for quick edits.
5. Test and commit, then open a PR and let Bugbot or a teammate review it.
6. Update your rules whenever you correct the agent twice for the same thing.

Troubleshooting

The agent ignores my rules

Check the rule type. A rule set to apply only to certain files will not load for other files. Make key rules “Always”, keep them short and put the most important line first.

“You’ve hit your usage limit”

The new plans share one weekly pool with Grok, so heavy Grok chat also drains Cursor. Switch to Auto, use Tab and Ctrl/Cmd+K for small jobs, wait for the weekly reset or move up a plan. On Premium, check the usage page for third-party model spend.

GPT models disappeared

OpenAI’s built-in access is ending on November 12, 2026. Pick Claude, Gemini, Grok or Composer, or add your own OpenAI key in Settings > Models.

Claude is not in the model list

On the new plans, Claude and Gemini start at Premium. Plus is Grok only.

The agent edits the wrong files

Name the files with @, ask for a plan first, and use a new chat per task. Add a .cursorignore for folders it should never touch.

Terminal commands hang

Interactive prompts (like “Are you sure? y/n”) block the agent. Tell it to use non-interactive flags (-y, --yes, CI=1), or cancel and run the command yourself.

Cursor feels slow or uses lots of RAM

Disable extensions you do not need, close old agent tabs and exclude large folders from indexing. Restarting the extension host (Command Palette > “Developer: Restart Extension Host”) fixes most freezes; Cursor 3.26 can resume agents stopped by an extension host crash.

Indexing never finishes

Very large repos or huge generated files slow it down. Add them to .cursorignore and click Resync in Settings > Indexing & Docs.

My Take

Cursor is still the easiest way to get a strong coding agent inside a familiar editor. The skill is not the shortcuts, it is the habit: plan first, give clear rules, keep chats short and review every diff.

The new X Subscription makes the $8 plan a cheap way in, but if you want Claude in your editor, budget for Premium plus API usage, or compare it with Claude Code pricing, which you can run in Cursor’s terminal anyway.

If you are new to Claude itself, start with how to use Claude AI.

FAQ

Is Cursor AI free?

Yes, there is a Free plan with a starter allowance of agent requests and no card needed. Paid plans start at $8 a month (Plus, Grok models only).

How do I use Cursor AI for beginners?

Install it, import your VS Code settings, open a project and try three things: accept Tab suggestions, select code and press Ctrl/Cmd+K to change it, and open the Agent with Ctrl/Cmd+I to ask for a feature. Always review the diff.

What is the difference between Cursor’s Agent, Plan and Ask modes?

Agent edits files and runs commands. Plan writes a step-by-step plan you approve first. Ask only answers questions and never changes code.

Can I use Claude in Cursor?

Yes. On the new plans Claude needs Premium ($30) or higher and is billed at API list price. Legacy Pro plans keep their existing model access.

Will GPT models still work in Cursor after November 12, 2026?

Built-in access is expected to end on that date. You can bring your own OpenAI API key for Chat and Agent, but not for Tab, Auto, cloud agents or the CLI.

Is my code safe in Cursor?

Turn on Privacy Mode so your code is not stored or used for training, keep secrets in .cursorignore and only add MCP servers and extensions you trust.

Do my VS Code extensions work in Cursor?

Most do, because Cursor is based on VS Code. You can import them on first launch.

What is the Cursor Agents Window?

A separate window added in Cursor 3 that shows and manages all your agents, local and cloud, with projects, task boards and worktrees.

Sources

• Cursor release notes (IDE 3.26 and 3.27, Agents Window, CLI, SDK)
• Cursor CLI changelog
• Teslarati: The X Subscription combines Grok, Cursor and X (Oct 13, 2026)
• Fello AI: Cursor pricing and legacy plans (updated Nov 4, 2026)
• Tom’s Guide: OpenAI is leaving Cursor in November
• DataCamp: What is Cursor 3
• Cursor release tracker: live apps (Nov 5) and voice chat (Nov 6)