Codex CLI Guide

Codex CLI is OpenAI's terminal-based coding agent, similar to Claude Code but powered by GPT models.

Sandboxing is on by default. Expect restricted filesystem + network access until you explicitly lift restrictions per task.

Setup

# npm
npm install -g @openai/codex

# Homebrew
brew install --cask codex

export OPENAI_API_KEY=sk-...
codex

How It Works

Codex CLI reads your project structure, understands your codebase, and can:

Generate and edit files across your project
Run shell commands and interpret results
Create git commits with meaningful messages
Execute test suites and fix failures

Key Differences from Claude Code

Feature	Codex CLI	Claude Code
Model	OpenAI current flagship	Claude Sonnet/Opus
Provider	OpenAI	Anthropic
MCP support	No (as of 2026-04-17) *	Yes (native)
Background agents	No	Yes
Config file	None (use `AGENTS.md`)	CLAUDE.md
Sandboxing	Sandboxed by default	Configurable

* Verified 2026-04-17 against the github.com/openai/codex README, which does not mention MCP support. Re-check OpenAI's platform docs for updates.

Autonomy Levels

Codex CLI offers three modes:

Suggest — proposes changes, waits for approval
Auto Edit — applies file changes automatically, asks for commands
Full Auto — runs everything autonomously in sandbox

Best For

Developers already in the OpenAI ecosystem
Projects that benefit from GPT-4.1's strengths
Teams wanting sandboxed-by-default execution

Windsurf Guide

Windsurf IDE with Cascade agent for flow-based AI coding.

Zed Guide

A fast, collaborative, AI-native code editor with Agent Panel, Inline Assistant, and Edit Prediction.

On this page

Setup How It Works Key Differences from Claude Code Autonomy Levels Best For