Codex

📖 4 min read openaicodexcodingagenticmcp

Complete guide to OpenAI Codex — the agentic coding platform across Desktop, IDE, CLI, Web, and CI/CD. Setup, workflows, AGENTS.md, Skills, MCP, Hooks, Computer Use, and comparison with Cursor/Claude Code.

Key Takeaways

Codex runs on 7 surfaces: Desktop app, IDE extension, CLI, Web (cloud), Chrome extension, GitHub Action, and Codex SDK
AGENTS.md = OpenAI's equivalent of CLAUDE.md. Skills = same SKILL.md standard as Anthropic. MCP = same open protocol
Codex supports Subagents, Automations (scheduled tasks), Worktrees (parallel branches), and Computer Use
Integrates with GitHub, Slack, Linear for end-to-end developer workflows

Codex is OpenAI’s agentic coding platform. It reads your entire codebase, edits files across multiple files, runs terminal commands, and integrates with your development workflow — from Desktop to CI/CD.

Installation — 7 Surfaces

Surface	Best For	Key Feature
Desktop App	Visual workflow, multi-session	Computer Use, Appshots, Automations, Worktrees
IDE Extension	VS Code integration	Inline diffs, slash commands, @-mentions
CLI	Terminal-first, scripting	`codex "prompt"` for quick tasks, non-interactive mode
Web (Cloud)	No setup, browser	Cloud environments, internet access
Chrome Extension	Browser-based actions	Web automation, scraping, testing
GitHub Action	CI/CD automation	Automated PR review, issue triage
Codex SDK	Build custom agents	Full programmatic control

# Install Codex CLI
pip install openai-codex

# Start in any project
cd your-project
codex

# Non-interactive mode for CI/CD
codex --non-interactive "review this PR for security issues"

Core Capabilities

AGENTS.md — Project Instructions

Like Anthropic’s CLAUDE.md, Codex reads AGENTS.md from your project root at session start:

## Code Style
- Use TypeScript strict mode
- Prefer functional patterns over classes
- Tests colocated with source in __tests__/

## Architecture
- API routes in src/api/
- Business logic in src/services/
- Database queries in src/db/

Skills — Same SKILL.md Standard

# Project skills
.codex/skills/
  generate-migration/
    SKILL.md
  review-pr/
    SKILL.md

Codex supports the same SKILL.md open standard as Claude Code. Skills are portable across 32+ tools. See Agent Skills for details.

MCP — Model Context Protocol

{
  "mcpServers": {
    "github": {
      "command": "npx", "args": ["-y", "@modelcontextprotocol/server-github"]
    },
    "slack": {
      "command": "npx", "args": ["-y", "@modelcontextprotocol/server-slack"]
    }
  }
}

See MCP & Integrations for the full guide.

Hooks

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{"type": "command", "command": "prettier --write ${CODEX_TOOL_INPUT_FILE}"}]
    }],
    "PreToolUse": [{
      "matcher": "Bash(git commit*)",
      "hooks": [{"type": "command", "command": "npm run lint && npm test"}]
    }]
  }
}

Subagents — Parallel Task Execution

codex "audit all services for vulnerabilities, checking each in parallel with subagents"

A lead agent coordinates, assigns subtasks, and merges results. See the Agents SDK for building custom subagent architectures.

Computer Use

Codex can control your browser and desktop applications — navigate websites, fill forms, test UIs:

codex "go to github.com/trending, find the top Python repo, clone it, and analyze the codebase for security issues"

Available on Desktop, CLI, and Web surfaces.

Automations & Worktrees

Automations: Scheduled tasks that run on your machine or in the cloud — morning PR reviews, nightly dependency audits
Worktrees: Parallel git branches that Codex works on simultaneously — isolate experiments without cluttering your workspace

Codex vs Cursor vs Claude Code

Feature	Codex	Cursor	Claude Code
Primary interface	Desktop app + IDE + CLI + Web	IDE fork (VSCode-based)	Terminal CLI + IDE + Desktop
Models	GPT-5.5 / 5.4 / 5.4 mini	Claude Sonnet + GPT	Claude Opus 4.7 / Sonnet 4.6
Project instructions	AGENTS.md	`.cursorrules`	CLAUDE.md
Skills standard	SKILL.md (same as Claude)	SKILL.md	SKILL.md
MCP support	Yes (same protocol)	Yes	Yes
Computer Use	Yes (Desktop, CLI, Web)	No	Cowork (research preview)
Subagents	Yes	No	Yes
Scheduled tasks	Automations	No	Routines
CI/CD	GitHub Action, Codex SDK	None	GitHub Actions
Pricing	ChatGPT Plus/Pro or API tokens	$20-40/mo	Pro/Max or API tokens
Unique strengths	GPT-5.5 reasoning, 7 surfaces, computer use, Commerce/Ads platform	Largest IDE market share	Best reasoning, 1M context, Agent Skills pioneer

Integrations

Integration	What It Does
GitHub	Auto PR review, issue triage, repo management
Slack	`@Codex` mention — bug reports become PRs
Linear	Issue → code fix pipeline

Pricing

Plan	What You Get
ChatGPT Plus ($20/mo)	Codex access on 1 surface
ChatGPT Pro ($200/mo)	Codex across all surfaces, extended usage
API (Platform)	Pay-per-token via platform.openai.com
Enterprise	Custom plans, admin controls, SSO