Overview
Cursor is an AI-powered code editor forked from VS Code that integrates frontier AI models directly into the coding experience. It offers AI-driven code completions, chat, agent mode for multi-file changes, and cloud-based coding agents. It supports every VS Code extension and keybinding.
Our take
Cursor is an editor built around AI assistance rather than an editor with AI added, and that difference shows in how much of your codebase it can see and act on. For many developers it has become the default. The adjustment it requires is reviewing generated changes carefully, because it will happily make large ones quickly.
Worth it if
Developers who want deep AI assistance across a whole repository rather than line completion.
Look elsewhere if
Codebases that cannot leave your network without an enterprise arrangement covering it.
✨ Features
AI Agent Mode
Autonomous coding agent that handles multi-file edits, refactors, and feature implementation.
Tab Completions
Intelligent code completions and next-edit suggestions powered by frontier models.
Cloud Agents
Run coding tasks in the cloud with background agents and automations.
MCP & Skills
Extend with Model Context Protocol servers, custom skills, and hooks.
⚖️ Pros & Cons
Pros
Cons
💰 Pricing
Hobby
Free
- No credit card required
- Limited Agent requests
- Limited Tab completions
Individual
$20 /mo
- Extended Agent limits
- Access to frontier models
- MCPs, skills, and hooks
- Cloud agents
- Bugbot on usage-based billing
Teams
$40 /user/mo
- Centralized billing and admin
- Team marketplace for rules and plugins
- Agentic code reviews with Bugbot
- Usage analytics
- SAML/OIDC SSO
Enterprise
Custom
- Pooled usage across organization
- Invoice/PO billing and SCIM
- Repository, model, and MCP controls
- Audit logs and service accounts
- Priority support
Categories
❓ Frequently Asked Questions
Is my code sent to a server?
By default context goes to model providers, with privacy modes and enterprise arrangements available that change retention and training terms. If your repository is under NDA or contains anything regulated, settle this before evaluating anything else — it is the term that decides whether the tool is usable at all.
How should code review change when using it?
Read generated diffs as carefully as a stranger's pull request, because the errors have a different shape. Generated code is usually idiomatic and syntactically clean, so it slips past review scanning for obvious mistakes. The recurring issues are outdated APIs, missing error paths and hallucinated dependencies.