Cursor Guide — Tell AI to Build, No Code

Cursor is a Code Editor with built-in AI — think of it as your "assistant who writes code for you." Just say what you want, and AI handles the rest.

This guide is organized into 9 categories with 38 topics, arranged from basics to advanced — read it all or jump to whichever category interests you.

1. What is Cursor — Starting from Zero

1.1 What is Cursor? Why Switch from VS Code?

Cursor is a Code Editor forked from VS Code — it looks exactly the same, all your existing extensions work, but AI is baked in from day one.

The reason to pick Cursor? It transforms the workflow from "sitting down to write code" to "sitting down to direct AI" — describe what you want, AI writes it, just review and hit Accept.

1.2 Installation and First-Time Setup

Download Cursor from cursor.com — available for Mac, Windows, and Linux. Once installed, open your project right away. All VS Code extensions work out of the box.

Things to set up after installation:

1. Choose an AI Model — Start with Auto Mode (Cursor picks the best model for you)
2. Privacy Mode — If working with client data, enable Privacy Mode so your code isn't sent to train AI
3. Rules for AI — Add personal rules like "Reply in English" or "Use Tailwind CSS only"

1.3 The Only 3 Shortcuts You Need to Remember

Cursor has tons of features, but memorizing just these 3 shortcuts covers 80% of what you'll need:

1.4 Tab Completion — Let AI Auto-Complete Code

Beyond the 3 main shortcuts, Cursor has a Tab Completion feature running in the background at all times — as soon as you start typing something, AI suggests code it thinks you'll type next. Press Tab to accept, or just keep typing if you don't want it.

1.5 Checkpoints — Time Travel to Undo Changes

Every time you chat with AI and it edits code, Cursor automatically creates a Checkpoint — like a "Save Point" in a game. If AI breaks something or you don't like the result, jump back to the previous Checkpoint instantly.

Think of it as "Git for AI conversations" — no need to worry about AI breaking things, because you can always revert.

2. @ Symbols — Teaching AI to Understand Your Project

The biggest problem with AI is "it doesn't know what's in your project" — @ Symbols are how you tell AI "go read this file, this folder, this website, before answering."

The biggest problem with AI is "it doesn't know what's in your project"

2.1 @Files & @Folders — Specify Relevant Files

@Files — Point AI to a specific file to read. AI will see 100% of that file's content.

@Folders — Point to an entire folder. Great for structural questions.

2.2 @Codebase — Search Across the Entire Project

@Codebase lets AI perform a Semantic Search across the entire project — not just matching words, but understanding the "meaning" of code.

2.3 @Docs — Pull in the Latest API Documentation

AI was trained on older data — @Docs fixes this by letting you add URLs to the latest API documentation. AI reads from the actual docs instead of guessing.

How to add @Docs:

Start here:

1. Press Cmd+Shift+P and type "Cursor: Add Documentation"
2. You enter the URL of the API docs you need
3. Wait for Cursor to finish indexing
4. Use @Docs in chat to pull information from those docs

2.4 @Git — View Edit History

@Git gives AI access to the project's Git history — view recent diffs, see who changed what and when. Handy for reviewing code before committing.

2.5 @Web — Search the Internet

@Web lets AI search the web — perfect for unfamiliar error messages or when you need the latest info that AI doesn't know yet.

2.6 Notepads — Persistent Notes Across Sessions

Notepads are permanent text blocks that AI can read at any time — unlike chat which disappears when you start a new one, Notepads stick around until you delete them.

Use Notepads to store frequently needed info like:

• Database schema
• Existing API endpoints
• Company rules or coding standards

3. Three Work Modes — Chat, Plan, Agent

Cursor has 3 main modes that each work differently — picking the right mode means 10x better results.

3.1 Chat Mode (Cmd+L) — Q&A + Mermaid Diagrams

Chat Mode is the most basic mode — ask anything, AI answers. Get explanations, debug, write new code.

Standout feature: You can ask AI to draw Mermaid Diagrams right in the chat — see workflow diagrams visually without opening another app.

3.2 Plan Mode (Shift+Tab) — Let AI Be the "Architect" Before Building

Plan Mode is a game changer — instead of having AI write code immediately, tell it to analyze first, plan first, then execute.

How to enter Plan Mode:

1. Press Cmd+L or Cmd+I to open chat
2. Press Shift+Tab to switch to Plan Mode
3. AI will focus on "analysis" instead of "writing code"

3.3 Agent Mode (Cmd+I) — Tell AI to Work Across Multiple Files

Agent Mode is the most powerful mode — AI works autonomously, creates new files, edits existing ones, runs Terminal commands, does everything on its own until the job is done.

3.4 When to Use Which Mode — Quick Decision Guide

Great tools bloom into great results — keep building

4. Prompt Techniques — Getting AI to Deliver What You Actually Want

4.1 Principles of Writing Good Prompts

A good prompt needs 3 components:

4.2 The "AI Lead" Technique — Let AI Take the Lead First

This technique is really important — instead of spelling out every detail for AI, describe the goal broadly and let AI propose a plan first, then refine from there.

The benefit is that AI will suggest approaches that fit your codebase — not just paste a template from somewhere else.

3 Prompts for Scaffolding a New Project When starting a new project, combine Plan Mode + Agent Mode: EXAMPLE PROMPT — Scaffolding Create a new project structure for an E-commerce MVP: Frontend: Next.

4.3 Prompts for Scaffolding a New Project

When starting a new project, combine Plan Mode + Agent Mode:

4.4 Prompts for Refactoring Legacy Code

5. Choosing the Right AI Model for the Job

Cursor offers multiple AI Models to choose from — each one excels at different things. Picking the right one makes a huge difference in results.

5.1 Claude vs GPT vs Gemini vs DeepSeek

5.2 Model Selection by Task Type

5.3 Auto Mode — Let Cursor Choose Automatically

Not sure which model to use? Pick Auto Mode and Cursor will assess the task's complexity and choose the best model for you.

5.4 Model-Switching Strategy When Quality Drops

If one model starts giving worse answers — don't keep sending the same prompt. Switch to a different model instead.

6. .cursorrules — Create Rules AI Always Follows

The problem: every time a new chat opens, AI forgets everything about your project's technologies and rules — .cursorrules fixes this by creating a rule file that AI reads automatically every time.

6.1 What is .cursorrules and Why You Need It

The .cursorrules file is a "System Prompt" that gets sent along with every question you ask AI — whether using Cmd+K, Cmd+L, or Cmd+I.

M

The problem: every time a new chat opens, AI forgets everything about your project's technologies and rules

6.2 A Well-Structured File

A good .cursorrules file should include these sections:

6.3 Example .cursorrules for Next.js

Using Next.js? Here's a template ready to copy and use:

6.4 Enforce Testing & Edge Cases

You can force AI to write tests every time a new function is created — just add this rule to .cursorrules:

6.5 Security Guardrails

Prevent AI from writing code with vulnerabilities — add security rules to .cursorrules:

cursorrules: Add to .

6.6 Protected Zones — Prevent AI from Touching Critical Code

Got code that's off-limits? Like performance-critical sections or core logic:

7. Memory Bank — An External Brain So AI Remembers Across Sessions

7.1 The "Short-Term Memory" Problem

The biggest issue with AI in Cursor is Stateless Sessions — every time a new chat opens, AI forgets everything from previous conversations. You have to explain everything from scratch.

And when chats get too long, answer quality starts dropping because the "Context Window" can't handle all the information.

The fix: create a Memory Bank — an "external brain" that AI reads every time it starts working.

7.2 Setting Up .cursor/memory/ — The 3 Core Files

Create a .cursor/memory/ folder at the project root and add these 3 files:

7.3 Connecting Memory Bank to .cursorrules

Add this directive to .cursorrules so AI is forced to read the Memory Bank before starting work:

7.4 Session-End Ritual — Update Memory Before Wrapping Up

Before finishing each work session, tell AI to update the Memory Bank:

8. Working as a Team with Cursor

8.1 .cursor/plans/ — Save and Share Work Plans

Before starting a big task, use Plan Mode to create a work plan and save it in .cursor/plans/ — commit it to Git so the team can see and review together.

8.2 Phased Implementation — Break Work into Phases

Big tasks must be split into phases — complete one part and test it before moving to the next. If something breaks, you'll know exactly which phase caused it.

8.3 Reference Implementation — Create Examples for AI to Follow

A technique that works really well for repetitive tasks like multiple migrations:

1. Get the first file 100% right by yourself (or AI + your review)
2. Commit that file
3. Tell AI "use this file as the example, do the rest following the same pattern"

AI learns from "real examples" way more accurately than reading rules as text.

8.4 Code Review with Agent Mode + @Git

Before committing or creating a Pull Request, use Agent Mode + @Git to have AI review the code:

8.5 Custom Slash Commands

Create shortcut commands in Cursor for the team to share — place .mdc files in .cursor/rules/:

• /commit — AI drafts the commit message automatically
• /pr — AI summarizes the work as a Pull Request description
• /test — AI writes Unit Tests for the selected function

8.6 MCP — Connect AI to External Systems

MCP (Model Context Protocol) is a standard for connecting AI to external tools — like databases, APIs, and project management systems.

Real-world examples that actually work:

• Connect AI to Supabase — ask AI "how many rows are in this table" and AI queries it directly
• Connect AI to GitHub — tell AI to create issues or review PRs
• Connect AI to Slack — tell AI to send work summaries to the team

9. Troubleshooting — When AI Won't Follow Instructions

9.1 AI Hallucination — Causes and Fixes

Hallucination is when AI "makes up answers" — using APIs that don't exist, suggesting functions that aren't in your codebase.

How to reduce Hallucination:

1. Use @Files to point to specific files (100% accurate — reads exact file) instead of @Codebase (~80% accurate — searches by embedding similarity)
2. Use @Docs to add latest API documentation so AI doesn't have to guess
3. Use Plan Mode to analyze before taking action
4. Have AI reference the Memory Bank before answering

9.2 AI Ignoring .cursorrules — 5 Fixes That Work

When AI starts ignoring rules in .cursorrules:

1. Start a new chat — chats that run too long cause context overflow, and AI starts forgetting rules
2. Use Memory Bank — move frequently changing info to .cursor/memory/
3. Trim .cursorrules — keep it under 10,000 words. If it's too long, split into reference files
4. Include real code examples — AI learns better from examples than from reading rules
5. Switch models — Claude Sonnet tends to follow complex rules better than other models

9.3 Context Overflow — Start Fresh + /summarize

When chats go beyond 20-30 messages, quality starts to drop:

• Summarize work with /summarize before closing the chat
• Start a fresh chat every time you begin new work
• Update activeContext.md before ending the session

9.4 Privacy & Security — Preventing Data Leaks

When working with sensitive data:

• Enable Privacy Mode — your code won't be sent to train AI
• Use .cursorrules — add a rule "never recommend putting secrets in code"
• Use DeepSeek (Local) — for maximum privacy
• Never paste real API Keys in chat — use placeholders and swap them in later

Wrap-Up — The Daily Workflow

Here's the workflow used with Cursor every single day:

This guide will be updated as Cursor rolls out new features — bookmark it and come back anytime.

Last updated: 2026-03-21

Related Articles

• 8 AI Bots Replace a Team — AI Operations Center
• AI Server Audit — One Person, 5 Roles, 3 Hours
• Claude Code Security — Permission, Sandbox, Hooks Guide

#Database #Security #Chatbot #Tutorial #CursorAI #NextJS #GPT #Server #Linux #AITools #PromptEngineering #Template