Set up Cursor: install, VS Code import, project rules, privacy, plans and MCP servers

Cursor is an AI code editor built on the VS Code codebase: the same layout, extensions and keyboard shortcuts, with an agent that reads your project, edits files and runs commands. Setting it up well takes ten minutes — install, bring over your VS Code setup, tell the agent your project's rules, decide what may leave your machine, pick a plan, and connect the tools you want the agent to use through MCP. Every step below follows Cursor's own documentation as we read it on 2026-11-07; Cursor changes quickly, so check the linked pages when something looks different.

1. Install

Download from cursor.com/download (version 3.29 on that date):

• macOS: Apple silicon (ARM64), Intel (x64) or a Universal build.
• Windows: x64 or ARM64, each as a "System" or a "User" installer.
• Linux: .deb, RPM or AppImage, for x64 or ARM64.

Cursor also has a terminal agent, the Cursor CLI, installed separately with curl https://cursor.com/install -fsS | bash on macOS, Linux and WSL, and started with the command agent — not cursor-agent, the name older tutorials use.

2. Bring over your VS Code setup

Cursor's VS Code migration page describes a one-click import: open Cursor Settings (Ctrl/Cmd + Shift + J), go to General > Account, and under VS Code Import click Import. It moves your extensions, themes, settings and keybindings. Check afterwards that the extensions you rely on — language servers, linters, formatters — are installed and enabled.

3. Write project rules

Rules tell the agent what it cannot see in the code: your conventions, commands, and things to never do. Cursor's rules documentation is precise about the format:

• Project rules live in .cursor/rules/ as .mdc files and are committed with the project. "A plain .md file in .cursor/rules is ignored."
• AGENTS.md in the project root (and subfolders) is the plain-Markdown alternative, and Cursor reads it too — useful if other agents on your team read the same file.
• User rules are personal and apply to every project; Team rules exist on Team and Enterprise plans.

A good first rule file is short: how to run the tests, the package manager, the code style, folders the agent must not touch, and "never commit secrets". Older guides mention a .cursorrules file in the project root; the current docs do not describe it, so use .cursor/rules or AGENTS.md for new projects.

4. Decide on Privacy Mode

Cursor's privacy help page says: "Privacy Mode ensures your code is never used for training by Cursor or other AI model providers." With it off, Cursor's data-use page says it "may use and store codebase data, prompts, editor actions, code snippets" to improve its features and train models.

Turn it on under Settings > General if your code is not yours alone to share — client work, anything under an NDA, anything your employer has a policy on. The defaults are not stated clearly for individual accounts, and Cursor's pages disagree for teams: one says Privacy Mode is on by default for all team members, another says for Enterprise teams. Check the setting rather than assume it.

5. Pick a plan

On 2026-11-07, Cursor's pricing and models and pricing docs describe a new model: Cursor is "sold as part of The X Subscription, one subscription for Grok, Cursor, and X".

Annual billing is $80, $300, $1,000 and $2,000 a year for Plus to Ultra. The included usage pool "refreshes weekly". If you specifically want Claude or GPT models in the agent, the cheapest plan that offers them is Premium, and those models draw on prepaid credit at the vendors' list prices. Earlier plans (Pro, Pro+ and the old Teams seats) are now legacy plans.

6. Connect tools with MCP

The Model Context Protocol lets the agent call outside tools — a database, an issue tracker, a design or image service — instead of only reading and writing files. Cursor's MCP docs support local servers (stdio) and remote ones over SSE or Streamable HTTP, configured in JSON:

• .cursor/mcp.json in the project, for that project (and anyone who clones it);
• ~/.cursor/mcp.json in your home folder, for every project.

A remote server is a url plus headers; Cursor resolves variables in command, args, env, url and headers, so keys can come from your environment instead of the file. Servers in Cursor's marketplace install with an Add to Cursor button and OAuth sign-in.

For example, to give the agent our image tools — rendering images from templates, taking screenshots of public pages, checking usage — add:

{
  "mcpServers": {
    "shareglint": {
      "url": "https://shareglint.com/mcp",
      "headers": { "Authorization": "Bearer ${env:OG_API_KEY}" }
    }
  }
}

and set OG_API_KEY in your environment before starting Cursor. The tools it adds are render_image, get_image, list_templates, get_template, create_signed_url, get_usage, generate_template, take_screenshot, auto_og_url, create_batch, get_batch. The Cursor MCP guide covers keeping the key out of git, tool approvals and a worked example that puts og:image tags into a Next.js app.

7. A sensible first week

• Review every change. The agent's diffs are proposals; read them as you would a colleague's pull request, especially in code that handles money, auth or data.
• Keep rules current. Each time the agent repeats a mistake, add one line to the rules instead of correcting it again.
• Approve tools that spend. Leave approval on for MCP tools that cost money or change things; allowlist only read-only ones.
• Commit often. Small commits make it easy to undo an agent's wrong turn.

Cursor or something else?

Cursor is an editor first; Claude Code is a terminal agent first, and the two are increasingly used together. If you are deciding between them, our Claude Code vs Cursor comparison puts the plans, models and workflows side by side, with sources.