Cursor

AI-first editor plus terminal agent with rules, MCP, permissions, headless automation, and multi-model workflows.

Type: IDE / CLI / agent

Best for: Daily editor-first coding with a capable terminal and automation path

Official docs: https://cursor.com/docs

Last verified: 2026-08-09

Status: Documentation verified

Verification scope: Official Cursor CLI installation, usage, headless mode, parameters, permissions, rules/MCP guidance, and the January 2026 CLI changelog were reviewed. No local Cursor install or login flow was executed in this repository.

60-Second Cheat Sheet

Need Command or file Notes
Install Cursor CLI `curl https://cursor.com/install -fsS \ bash` Official CLI installer
Start the terminal agent agent Current primary entrypoint
Backward-compatible entrypoint cursor-agent Still supported as an alias
Ask without coding agent "explain this repo; do not write code" Good low-risk first task
Run non-interactively agent -p "review these changes" Print/headless mode
Apply writes in scripts agent -p --force "update the docs" High-impact; review permissions first
List models agent models Current CLI model discovery
Resume latest chat agent resume Continue prior context
List chats agent ls Find previous sessions
Manage MCP agent mcp or /mcp CLI and in-session management
Manage rules /rules Create or edit rules from the CLI
Project rules .cursor/rules/ Same rules system as the IDE
Agent instructions AGENTS.md / CLAUDE.md CLI reads these at project root
User CLI permissions ~/.cursor/cli-config.json Global CLI configuration
Project CLI permissions .cursor/cli.json Repository-specific CLI configuration

Safe default: use interactive mode for implementation, explicitly say “do not write code” while planning, and treat --force as an automation permission decision rather than a convenience flag.

Five Prompts Worth Bookmarking

Understand a feature in the editor

Trace this feature from UI entry point to data/API layer. Name the key files, explain the data flow, and identify the safest place to change behavior. Do not edit yet.

Plan before editing

Read the relevant files and rules, then propose a focused implementation plan with files, tests, risks, and validation commands. Do not write code until I approve the plan.

Review a change

Review the current changes for correctness, regressions, security issues, and missing tests. Prioritize concrete findings and cite the exact files involved.

Fix a bug

Reproduce the bug from the existing code and tests, identify the root cause, implement the smallest fix, and run the narrowest relevant validation before broader checks.

Refactor with constraints

Refactor this module without changing public behavior. Follow the repository rules, preserve APIs, keep the diff focused, and show test/build results before suggesting extra cleanup.

What Cursor Is Best At

  • editor-first coding where context, diffs, terminal, and AI live in one workflow
  • teams that want both IDE agent behavior and terminal automation
  • project rules shared through .cursor/rules/
  • mixed instruction ecosystems using AGENTS.md or CLAUDE.md
  • multi-model workflows and model switching
  • MCP-backed integrations across editor and CLI

Installation

The current CLI installer is:

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

Verify the compatible CLI binary:

cursor-agent --version

Cursor announced agent as the primary CLI entrypoint in January 2026. cursor-agent remains backward compatible, so older docs and scripts may still use it.

The CLI installation docs explicitly cover macOS, Linux, and Windows through WSL. Native Windows users can still use the Cursor editor; validate your preferred terminal setup before standardizing CLI usage.

First Session

cd /path/to/your/repository
agent

Start with a read-only request:

Explain the repository structure, build command, test command, and one low-risk improvement. Do not write any code.

Cursor CLI asks for approval before terminal commands in interactive mode.

Daily CLI Commands

agent
agent "review the auth module"
agent -p "summarize the current diff"
agent models
agent ls
agent resume
agent mcp

The older cursor-agent spelling remains useful when following existing CLI reference pages:

cursor-agent -p "review these changes" --output-format text
cursor-agent mcp list
cursor-agent update

Useful in-session commands include:

/models
/rules
/mcp
/compress

Use the current CLI help for the exact command list installed on your machine.

Rules and Repository Instructions

Cursor CLI supports the same rules system as the IDE through:

.cursor/rules/

The CLI also reads project-root:

AGENTS.md
CLAUDE.md

For teams, pick a deliberate convention instead of maintaining the same instruction in three places.

A practical pattern is:

  • use .cursor/rules/ for Cursor-specific scoped rules
  • use AGENTS.md for cross-agent repository instructions
  • keep CLAUDE.md only when Claude-specific guidance is genuinely different

Permission Model

Cursor CLI documents permission tokens for:

  • Shell(commandBase)
  • Read(pathOrGlob)
  • Write(pathOrGlob)

Permissions can be configured globally in:

~/.cursor/cli-config.json

or per project in:

.cursor/cli.json

Deny rules take precedence over allow rules.

A useful baseline is to deny destructive shell commands, environment files, and private-key patterns even when normal source reads/writes are allowed.

Headless and Automation Mode

Use print mode for scripts and CI:

agent -p "review this repository"

Historically documented examples use cursor-agent -p; the current primary entrypoint is agent.

Writing from print/headless automation can require --force:

agent -p --force "update generated documentation and run validation"

Treat --force as a separate approval tier. Automation should pair it with explicit CLI deny rules and a tightly scoped repository/task.

MCP and External Tools

Cursor CLI can detect and use the MCP configuration shared with the editor. The current CLI also provides MCP management commands, including listing servers and tools.

Review every server for:

  • read versus write capabilities
  • authentication and token storage
  • network access
  • destructive operations
  • whether the same MCP should be enabled in both IDE and automated CLI contexts

Team Adoption Checklist

  • [ ] choose one primary repository instruction pattern
  • [ ] define global and project CLI permission defaults
  • [ ] deny secrets and destructive commands explicitly
  • [ ] pilot interactive editor/CLI usage before --force automation
  • [ ] approve MCP servers separately from Cursor itself
  • [ ] document whether WSL is the supported Windows CLI path

Security Notes

  • editor assistance becomes an execution surface when terminal tools or CLI automation are enabled
  • --force can remove meaningful confirmation steps
  • rule layering can create hidden or conflicting guidance
  • MCP expands the data/tool boundary
  • API keys for automation should never be committed to the repository

Alternatives

  • GitHub Copilot for GitHub-native editor, CLI, review, and repository workflows
  • Claude Code for terminal-first work with a strong permissions and project-memory model
  • OpenAI Codex for terminal-first OpenAI workflows and dedicated local review
  • Gemini CLI for a Google-backed terminal workflow with sandbox and context controls

See the Comparison Matrix for a side-by-side shortlist.

Verification Status

  • Status: Documentation verified
  • Last verified: 2026-08-09
  • Scope: CLI installation, current entrypoint, interactive/headless usage, rules, permissions, sessions, models, and MCP reviewed against official Cursor sources
  • Not locally tested: installer execution, sign-in flow, IDE behavior, model availability, and team-admin settings

Sources