Cursor vs VS Code: what the fork buys and what it costs

They are the same editor, one fork apart

Anysphere did not build an editor. It took the open-source VS Code codebase and built on top of it, which is why Cursor opens looking like the thing you already use, imports your settings.json, and honours the keybindings in your muscle memory. Nothing about the file tree, the command palette, or the terminal panel is new.

That lineage is the most useful fact in this comparison and the one most articles bury. It means the switch is close to free in learning cost, and it means the differences are a short, specific list rather than a philosophy.

What the fork buys you

Four things, in descending order of how much people who switched say they would miss them.

1. Tab. Ordinary completion predicts tokens where your cursor is. Tab predicts your next edit, which is usually somewhere else in the file and usually the mechanical consequence of the change you just made. Rename a field and Tab walks you through every call site. This is not a better autocomplete; it is a different interaction, and it is the only item on this list VS Code has no answer to.
2. An agent that lives in the editor. Multi-file edits with the diffs inline, commands run and their output read back. VS Code has this now too, through agent mode, so the gap here is polish rather than capability.
3. Parallel agents with isolation built in. Cursor’s Agents Window runs several agents at once, each in its own git worktree, so they cannot overwrite each other. That is a real feature and it is the one most people underuse. The same shape is available outside any editor: see running Claude Code sessions in parallel.
4. Cloud agents and Bugbot. Work that continues when your laptop is shut, plus automated review on pull requests. Copilot has an equivalent on the GitHub side.

What the fork costs you, and it is not the $20

Microsoft’s marketplace terms restrict the extension marketplace to Microsoft’s own products, so forks cannot use it. Cursor therefore pulls from Open VSX, and Anysphere ships audited replacements for some of what is missing. Most extensions are on Open VSX and install without incident. The exceptions are the ones Microsoft wrote and kept proprietary, and in April 2025 those started enforcing the restriction in the extension binaries rather than only in the terms.

• You also give up being on the mainline. A fork tracks upstream on its own schedule, so VS Code releases arrive in Cursor later, and occasionally differently.
• And you give up the price of free. VS Code costs nothing and is MIT licensed. Whatever you conclude about Tab, the baseline you are comparing against is genuinely $0.
• What you do not give up is your configuration. Settings, keybindings, themes, and non-Microsoft extensions carry across, which is why the switch feels cheap right up until you need one of the five rows above.

The price comparison people get wrong

"$20 versus free" is not the comparison. VS Code is free, but the AI in it is a Copilot subscription, and the honest matchup is Cursor Pro against VS Code plus Copilot. On that basis Cursor is twice the price of the nearest paid equivalent, not infinitely more expensive than nothing.

The last row is worth more attention than the prices above it. Cursor keeps working and grows your invoice; Copilot caps your invoice and stops your premium requests. Neither is better, but one of them can surprise you on the 28th of the month. The same split is covered in more detail in Cursor vs GitHub Copilot, and the per-plan mechanics are in Cursor pricing.

VS Code is no longer the AI-free option

The comparison that circulated through 2025 assumed VS Code was a plain editor you bolted a completion extension onto. Microsoft moved the agent into the product. As of August 2026 VS Code documents agent mode, chat and inline chat, custom instructions, MCP server support, and bring-your-own-key for models, all in the editor itself.

• Agent mode plans across files, runs tests, and reports back, with approval controls before it writes.
• MCP servers connect the same tools you would wire into any other agent, so the integration story is not a Cursor exclusive. See what agentic coding actually is for what the loop is doing underneath.
• BYOK means you can point it at your own provider key rather than buying capacity twice.
• Copilot Free is a real free tier, not a trial: 2,000 completions and 50 chat requests a month, indefinitely.

When neither editor is the deciding purchase

Both of these are buffer-first tools: the unit of work is a file you are looking at, and the AI helps you while you look at it. A large amount of 2026 agent work is not shaped like that. You describe an outcome, the agent works for twenty minutes, and you read a diff. For that shape, the editor is where you review, not where the work happens, and paying $20 for better prediction inside it is optimising the wrong half of your day.

• Claude Code and Codex run in a terminal, against any editor, and they are already included in subscriptions many people hold: Claude Pro and ChatGPT Plus both start at $20/mo. If you pay for either, you own a capable agent whether or not you ever open Cursor.
• They run headless, so the same agent works in CI, in a git hook, and on a schedule. No editor does that.
• They parallelise cleanly with git worktrees, which is the same isolation Cursor’s Agents Window gives you, available without adopting a fork.
• Your editor choice becomes free again. Plain VS Code, Cursor, Neovim, JetBrains: the agent does not care, because it is not in the editor.

If you want the editor-versus-agent question in its own right rather than as a footnote here, Cursor vs Claude Code is the page for it.

The honest call

Questions people ask

Keep reading

Continue through Head to head with Cursor vs Windsurf in 2026: Windsurf is now Devin Desktop.