Cursor is still easiest to explain as a code editor with an agent built into the workflow. The important update is that the editor is no longer the whole product. Current Cursor also includes cloud agents, automations, subagents, hooks, skills, Bugbot, Security Agents, and a terminal interface.

Start with the desktop application because that is where the everyday feedback loop is strongest.

Install the Desktop Application

Download Cursor from cursor.com. During onboarding, Cursor can import extensions, settings, keybindings, and themes from Visual Studio Code. That migration is useful because Cursor is familiar enough that the editor should not be the new thing you are learning. The agent workflow should be.

After installation, check these settings before you hand it a real repository:

• Enable Privacy Mode if the repository contains proprietary code.
• Decide whether model requests can use premium or Max Mode models.
• Add User Rules only for preferences that should follow you across every repository.
• Keep repository-specific rules in .cursor/rules/*.mdc.
• Configure trusted MCP servers in .cursor/mcp.json or ~/.cursor/mcp.json.

The rule of thumb is simple: personal defaults go in user settings, repository behavior goes in the repository, and external tools should be explicit enough that a teammate can audit them.

Install the Command Line Interface

Cursor also ships a terminal agent. Install it with the official installer:

curl https://cursor.com/install -fsS | bash

On Windows, use PowerShell:

irm 'https://cursor.com/install?win32=true' | iex

The command is agent, not cursor. Run agent inside a repository for an interactive session, or use print mode for automation:

agent -p "Review the staged changes for correctness."

File modification from print mode requires an explicit flag such as --force or --yolo, which is a name that is at least honest about the life choice you are making. Treat that as a real permission decision, not a convenience switch.

First Project Setup

For a serious repository, create the project-level files before asking the agent to edit code:

.cursor/
  rules/
    architecture.mdc
    testing.mdc
  mcp.json

Add skills only when a workflow needs more structure than a rule can provide:

.cursor/
  skills/
    release-checklist/
      SKILL.md

If the repository already has AGENTS.md, Cursor will read it as an instruction source. I still prefer putting Cursor-specific behavior in Cursor files so the reader can tell which tool owns which instruction.

Verification Checklist

Before you trust a new setup, ask Cursor to perform a read-only orientation:

Read this repository and summarize the build, test, lint, and release commands.
Do not edit files. Cite the files you used.

Then compare its answer to the repository scripts. If it invents commands, fix the rules or context before you use Agent mode for changes.