skip to content

Codex CLI

8 cheat sheets

Codex CLI#

OpenAI’s open-source agentic coding CLI (github.com/openai/codex). Runs GPT models directly in your terminal, edits code, executes commands, and supports OS-native sandboxing (Seatbelt on macOS, bwrap+seccomp on Linux). Configurable via ~/.codex/config.toml with profiles, MCP servers, and an experimental hooks system.

What’s in this section#

PageTopics
OverviewInstall, auth (ChatGPT plan vs API key), default model, session modes
Configurationconfig.toml schema — model, approval policy, sandbox mode, profiles, env
Slash CommandsFull reference: /init, /model, /plan, /fork, /resume, /mcp, /diff, /quit, …
Approvals & SandboxApproval policies, sandbox modes, platform mechanics, —full-auto
MCP Servers[mcp_servers.*] config, codex mcp subcommand, runtime /mcp
Tips & AGENTS.mdAGENTS.md authoring, profiles, hooks, codex exec, codex cloud

Sections#

  • Overview — Install, authenticate, and understand interactive vs exec modes.
  • Configuration — Every key in ~/.codex/config.toml explained.
  • Slash Commands — The full interactive slash-command surface.
  • Approvals & Sandbox — Control what the agent can do and how it’s sandboxed.
  • MCP Servers — Connect external tools via the Model Context Protocol.
  • Tips & AGENTS.md — Advanced workflows: AGENTS.md, profiles, hooks, non-interactive use.

[!TIP] Start with Overview and Configuration, then read Approvals & Sandbox before running in --full-auto mode.

Codex CLI Overview

Install and use OpenAI's Codex CLI — the open-source agentic coding tool. Covers authentication, interactive mode, headless exec, ~/.codex/ layout, and AGENTS.md project memory.

May 25, 2026
codex openai cli agentic install headless agents-md

Codex Configuration

Complete config.toml reference for Codex CLI — model selection, approval policies, sandbox modes, profiles, MCP server stubs, shell environment, notify hooks, and project trust levels.

May 25, 2026
codex openai config toml configuration settings profiles

codex exec — Non-Interactive Mode

Complete reference for codex exec — scripted invocation, stdin piping, JSON event streams, output formats, exit codes, timeouts, and automation patterns for CI, git hooks, and shell pipelines.

May 25, 2026
codex openai exec non-interactive cli automation ci scripting

Codex Slash Commands

Complete reference for every built-in Codex CLI slash command — session control, model switching, sandboxing, MCP management, and TUI utilities.

May 25, 2026
codex openai slash-commands cli reference tui

Approvals & Sandboxing

Control what Codex CLI can do — approval policies (untrusted, on-request, never), sandbox modes (read-only, workspace-write, danger-full-access), OS-native sandbox mechanics, and --full-auto.

May 25, 2026
codex openai sandbox approvals security seatbelt bwrap seccomp permissions

Codex MCP Servers

Add and configure Model Context Protocol (MCP) servers in Codex CLI — config.toml [mcp_servers.*] reference, the codex mcp subcommand, per-server tool approval modes, and the /mcp runtime command.

May 25, 2026
codex openai mcp model-context-protocol tools extensions

Codex Tips & AGENTS.md

Advanced Codex CLI workflows — AGENTS.md authoring, profiles, experimental hooks, notify callbacks, non-interactive exec scripting, session resume and fork, and codex cloud.

May 25, 2026
codex openai tips workflow agents-md hooks profiles productivity

Codex Sub-Agents & Task Delegation

How Codex CLI supports delegating to child agents — the experimental /agent system, codex exec as a sub-agent pattern, MCP-mediated agent-to-agent calls, isolation boundaries, and orchestration patterns.

May 25, 2026
codex openai subagents agents delegation orchestration task-tool multi-agent