How to Set Up Cursor AI IDE

Cursor is the AI-native code editor — a VS Code fork rebuilt around an in-editor agent. In April 2026 it’s the market leader for editor-based AI coding, reportedly closing a ~$50B valuation round and sitting around $2B ARR. For fast, in-flow coding — not long autonomous jobs — this is the tool most developers have settled on.

This guide gets you from download to a productive setup in under an hour, and then explains the credit-pricing gotcha that burned a lot of people in 2025.

Step 1 — Install

Download from cursor.com. macOS, Linux, and Windows builds. If you’re coming from VS Code, your extensions and settings will import on first launch — Cursor is a fork, so 95% of what you know carries over.

Step 2 — Sign in

Create an account and pick a plan. Free has limited requests, Pro is the default paid tier, Business/Enterprise add SSO and privacy guarantees. If your employer has strict data-handling rules, confirm “Privacy Mode” is on before you open source code.

Step 3 — Index your repo

Open a project folder and Cursor will offer to index it. Accept. Indexing embeds your code so the agent can retrieve relevant files on demand instead of only seeing what’s currently open. For monorepos, indexing takes a few minutes the first time and then updates incrementally.

Step 4 — Learn the three modes

Cursor has three main agent surfaces. Know which one to reach for:

• Inline edit (⌘K / Ctrl+K) — ask the agent to change the current selection. Best for small edits, renaming, “add types here.”
• Chat pane (⌘L / Ctrl+L) — conversational Q&A over the repo. Best for “where is X,” “explain this function.”
• Agent mode — multi-file autonomous edits. Best for “add a new endpoint end-to-end,” “implement this ticket.” Approve each change before it writes.

Step 5 — Pick models per mode

Cursor gives you a model picker. In practice:

• Inline edits: a fast Sonnet-class model — latency matters, not depth.
• Agent mode: Claude Opus 4.7 or GPT-5 class — quality matters, latency less.
• Chat: whichever handles your codebase best. Test a few; switch by task.

Step 6 — Rules file

Create .cursorrules at the repo root. This is Cursor’s version of a CLAUDE.md — persistent instructions the agent reads every session. Put style, conventions, and “never do this” rules here. Short and specific beats long and polite.

- TypeScript strict. No 'any' unless commented.
- Keep React components under 300 lines; split if larger.
- Prefer server components; only use 'use client' when needed.
- Don't modify files under /generated.

Step 7 — The credit-pricing gotcha

Cursor moved to credit-based pricing in mid-2025. The trap: a single long agent run can burn through your monthly credits in one afternoon (there’s a well-circulated story of a team depleting an annual subscription in a day). Protect yourself:

• Set a hard monthly spend cap in the account settings.
• Keep agent-mode tasks small. If a task needs hours of autonomy, hand it to Claude Code — see our Claude Code setup guide.
• For routine edits, use inline edit with a cheaper model. Save agent-mode for the hard cases.

Step 8 — Keyboard shortcuts worth memorising

• ⌘K — inline edit.
• ⌘L — chat pane.
• ⌘I — agent composer.
• ⌘⇧L — add current file to chat context.
• @ in chat — reference a file, symbol, or doc.

The workflow most people converge on

Cursor for the tight in-editor loop — small edits, refactors, quick explanations. Claude Code for anything that needs hours of reasoning or a full repo sweep. ChatGPT (web) for one-off questions you don’t want in your editor context. That stack — plus our token counter to estimate what everything costs — is the most common setup among shipping developers this year.