> cursor cli

Cursor's agent without the editor — cursor-agent runs interactive or headless, resumes sessions, follows your rules, and plugs into CI

getting started sessions modes & models print mode & automation permissions & sandbox worktrees & cloud rules & mcp interface & shortcuts

// getting started

7 commands

Install the Cursor CLI on macOS or Linux

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

Install the Cursor CLI on Windows

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

Add cursor-agent to your PATH

$ echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc

$ cursor-agent login

Authenticate with an API key for scripts

$ export CURSOR_API_KEY=key_your_token_here

Update the CLI to the latest version

$ cursor-agent update

Sign out and clear stored credentials

$ cursor-agent logout

// sessions

5 commands

Start an interactive agent session

$ cursor-agent

Resume the most recent conversation

$ cursor-agent resume

Browse past chats and pick one to resume

$ cursor-agent ls

Resume a specific conversation by ID

$ cursor-agent --resume="chat-id-here"

Start a fresh chat or fork the current one

$ /clear

// modes & models

5 commands

Plan an approach before writing any code

$ cursor-agent --plan "refactor the payment module"

Ask questions about code without edits

$ cursor-agent --mode=ask "how does caching work in this repo?"

Switch modes and debug during a session

$ /debug

Choose which model the agent uses

$ cursor-agent --list-models

Switch models from inside a chat

$ /model

// print mode & automation

6 commands

Get a scriptable answer in print mode

$ cursor-agent -p "list all TODO comments in this repo"

Pipe a diff or logs into the agent

$ git diff | cursor-agent -p "summarize these changes"

Get structured JSON output for parsing

$ cursor-agent -p "review auth.ts for security issues" --output-format json

Stream the agent's output as it works

$ cursor-agent -p "add tests for utils.ts" --output-format stream-json

Let the agent apply changes without prompts

$ cursor-agent -p "update CHANGELOG.md for the latest release" --force

Run the agent in a GitHub Actions workflow

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

// permissions & sandbox

4 commands

Allow or deny specific tools and commands

$ { "permissions": { "allow": ["Shell(npm)", "Shell(git)", "Read(src/**)"], "deny": ["Shell(rm)", "Read(.env*)", "Write(**/.env*)"] } }

Run commands inside a sandbox

$ cursor-agent --sandbox enabled "run the build"

Scope which paths the sandbox can touch

$ cursor-agent -p "refactor the module" --allow-paths "src,tests" --readonly-paths "config" --blocked-patterns "*.env"

Skip trust and approval prompts in automation

$ cursor-agent -p "deploy the preview" --trust

// worktrees & cloud

4 commands

Work in an isolated git worktree

$ cursor-agent -w feature-auth

Base a worktree on a specific branch

$ cursor-agent -w hotfix --worktree-base main

Run setup scripts when a worktree is created

$ { "setup": ["npm install", "cp .env.example .env"] }

Hand a long task off to a Cloud Agent

$ & refactor the auth module and add comprehensive tests

// rules & mcp

5 commands

Give the agent project rules and context

$ mkdir -p .cursor/rules

List configured MCP servers

$ cursor-agent mcp list

Inspect the tools an MCP server exposes

$ cursor-agent mcp list-tools linear

Enable, disable, or log in to an MCP server

$ cursor-agent mcp login linear

Configure a new MCP server

$ { "mcpServers": { "linear": { "command": "npx", "args": ["-y", "linear-mcp-server"], "env": { "API_KEY": "value" } } } }

// interface & shortcuts

6 commands

Discover slash commands and get help

$ /help

Add files and folders to the context

$ @src/auth.ts

Review the agent's changes before accepting

Ctrl+R

Insert a newline and exit the CLI cleanly

$ /setup-terminal

Run shell commands from inside a session

$ /shell

Tune the CLI with a config file

$ { "version": 1, "editor": { "vimMode": true }, "notifications": true }