Cursor: review
Cursor: ReviewInstall & set upNewsTechnical tutorial
Current as of September 2026, compiled from our own notes and news archive; text machine-generated and editorially checked. Check versions and commands in the official documentation.
Cursor is a code editor based on VS Code with a built-in AI agent. The agent explains your codebase, plans and builds features, fixes bugs and reviews changes. This review looks at who it is for, what it does well and less well, and what it costs (as of September 2026).
What it is and who it is for
Cursor feels immediately familiar to every VS Code user: you bring your extensions, themes and keyboard shortcuts with you. The AI is throughout the entire editor: smart tab completion, an agent panel (Cmd/Ctrl+I), a Plan Mode for larger tasks, cloud agents that work in the background, Bugbot for automatic code reviews, and a command-line agent for scripts and CI.
It suits developers who want to work graphically, review changes visually and be able to switch between models from different providers per task.
How it works
You open a project, give the agent a task and see the proposed changes as a diff that you accept or reject. In Plan Mode (Shift+Tab), the agent first examines the relevant files, asks questions and proposes a plan before writing anything.
You provide instructions via project rules (.mdc files in .cursor/rules/ with glob scoping), via AGENTS.md, via user rules, or for teams via centrally managed rules. You connect external tools via MCP in .cursor/mcp.json.
Strengths
- The best AI experience in a graphical editor. Tab completion, inline editing and the agent work seamlessly together, and every change can be assessed as a diff.
- Choice of many models. In addition to Cursor's own Composer 2.5 and the Grok models, you can use Claude, GPT and Gemini. Auto mode chooses per task, in the Cost, Balance or Intelligence modes.
- Fine-grained rules. With
globs, descriptions and manual@rules, you determine exactly which instructions apply when, and you shareAGENTS.mdwith other agents. - More than an editor. Cloud agents, Bugbot reviews on pull requests and a CLI with JSON output make Cursor also usable outside the editor.
- Privacy can be configured. Privacy Mode excludes training on your code, with zero data retention, and is on by default on Teams and Enterprise.
Weaknesses and risks
- Costs are hard to predict. The model with two credits (Cursor models versus other models at API rates) is flexible but not simple. If you work a lot with expensive external models, you quickly exceed the included usage.
- You have to enable privacy yourself on an individual plan. If you forget, the default policy applies to your code.
- Tied to the editor. If you prefer working in Neovim, JetBrains or purely in the terminal, you miss most of what Cursor does well; the CLI is a supplement, not a replacement.
- A VS Code fork. New VS Code features and some extensions arrive later or differently than in the original.
--forcein the CLI lets the agent write without confirmation; in CI that calls for a clean checkout and a human review.
Price
As of September 2026:
- Hobby: free, without a credit card, with limited agent requests and access to Composer.
- Pro: $20 per month, with included usage for Cursor and other models, unlimited tab completion and cloud agents.
- Pro Plus: $60 per month, with more generous limits.
- Ultra: $200 per month, with the highest limits.
- Teams: $40 per user per month; Enterprise on request.
Beyond your credit, you buy extra usage at API rates. Cursor is not open source.
Alternatives
- Claude Code: terminal-first, strong in hooks, permissions and CI, but only with Claude models.
- Codex: open-source CLI from OpenAI with a robust sandbox model.
- OpenCode: open-source terminal agent that works with many different providers.
- VS Code with Copilot: for those who want to stay in the original editor.
Conclusion
Cursor is the strongest choice for anyone who wants to use AI within a familiar graphical editor and likes to switch between models. The Pro plan is enough for most individual developers, as long as you use its own models or Auto in the Cost mode for routine work. If you mainly want to work unattended in the terminal or in CI, or want full control over the infrastructure, you are better off with Claude Code, Codex or OpenCode.