Cursor — Quick reference
VS Code reimagined with an AI agent that writes, runs, and fixes your code — describe it in plain English, review the diff, accept.
At a glance
What it is
A code editor built on VS Code with AI in every layer: Tab autocomplete, a file-aware chat sidebar, and an Agent that edits across files, runs the code, and iterates until it works.
Where it runs / pricing
Desktop app from cursor.com. Hobby free (no credit card) · Pro $20/mo · students get 1 free year of Pro.
Built on VS Code, so your familiar layout, shortcuts, extensions, and native Jupyter notebook support all carry over. It understands pandas, NumPy, matplotlib natively.
Setup & first edit
1 · Install
Download Cursor from cursor.com, install, and sign in. The Hobby plan is free, no credit card.
2 · Open a project
File → Open Folder so Cursor has a project to work in, then open the chat sidebar.
3 · First AI edit
Type your idea in plain English (any language). Cursor proposes the file + code as a diff — review and accept.
Create a Python file called analysis.py with a function that loads a CSV
with pandas and prints the first 5 rows. Add a short comment above each line.
Key commands & shortcuts
| Action | Mac | Windows / Linux |
|---|
| Open Chat / Agent sidebar | Cmd L | Ctrl L |
| Open Chat in sidebar (alt) | Cmd I | Ctrl I |
| Inline edit (edit selection in place) | Cmd K | Ctrl K |
| Accept a Tab suggestion | Tab | Tab |
| Add context (file/folder/symbol) | @ | @ |
| Accept / apply suggested changes | Cmd ⏎ | Ctrl ⏎ |
Cursor's docs list the Mac defaults verbatim; it uses VS Code keybindings as the baseline, so Windows/Linux follow the usual Cmd → Ctrl convention. Every binding is remappable in Keyboard Shortcuts settings.
Prompting patterns
Name the file, columns & rule
"…CSV expression_data.csv, gene names in column A, values in B–G, drop any gene with >20% missing, plot a clustered heatmap with seaborn." Specific scope keeps the agent on track.
Ask about your own code
"What does the function I just wrote do, and how would I call it on a column named TP53?" The chat reads your files, so answers are grounded in your project.
Paste the error, state the goal
"I got ModuleNotFoundError: No module named 'seaborn'. Fix it and make the script run end to end." Batch error + goal in one message.
Keep it small-scoped
Vague instructions let the agent drift; precise ones keep it accurate. One feature, named inputs, exact rules per request.
Context & modes
Tab
Autocomplete as you type — suggests whole lines and functions; press Tab to accept or keep typing to ignore. For moment-to-moment speed.
Chat sidebar
File-aware Q&A and small, located edits. Already knows your open files; @ a specific file/folder to pin context.
Agent mode
Autonomously reads the project, writes multi-file changes, runs the code, and iterates until green — you review each diff before it applies.
Model picker
Switch between frontier models per task — Claude, GPT, Gemini, or Auto. Heavier model for tricky reasoning, faster one for routine edits.
Beyond files: Cursor supports MCPs (connect the agent to external tools/data), plus skills, hooks, and cloud agents (extended on paid plans). Context is the single biggest quality lever.
Pricing & limits
| Plan | Price | Included usage |
|---|
| Hobby | Free | Limited Agent requests + limited Tab completions (tight — expect to hit them in one focused session) |
| Pro | $20/mo ($16/mo annual) | $20 included API usage + generous Auto/Composer usage; unlimited autocomplete |
| Pro+ | $60/mo | $70 included |
| Ultra | $200/mo | $400 included |
| Teams | $40 / user / mo | Central billing, team marketplace, Bugbot code reviews, shared-context cloud agents |
| Enterprise | Custom | Pooled usage, audit logs, service accounts |
Students Claim a free year of Pro at cursor.com/students — account email must match a valid .edu address; verify via Verify Status in account settings. Includes $20 usage/month, then auto-renews at regular pricing after one year (cancel anytime). Different models have different API costs, so model choice governs how fast included usage is spent.
Gotchas & tips
• The free Hobby tier runs out of requests quickly — meaningful daily use needs Pro ($20/mo).
• Cursor sends your code to cloud AI providers — not for confidential patient data or unpublished research without checking your institutional data policy.
• Agent mode can drift on vague instructions; precise prompts and small task scopes work better.
• Heavier than a plain text editor — slower startup on older laptops.
• The core loop: describe → run → paste the error → repeat. Batching the error and the goal in one message spends fewer Agent requests.