Back to Blog
BlogApril 23, 20262,564

OpenAI Codex Computer Use: Complete Step-by-Step Tutorial

OpenAI Codex Computer Use: Complete Step-by-Step Tutorial

Prerequisites

Before starting, ensure you have:

  • macOS (Computer Use is currently macOS-only)
  • Codex app installed and updated (download from OpenAI if needed)
  • ChatGPT Plus, Pro, Business, Enterprise, or Edu subscription
  • Basic familiarity with the Codex app interface

Note: Feature availability excludes EEA, UK, and Switzerland at launch.

Image

Step 1: Install the Codex App

  1. Visit the official OpenAI Codex page and download the latest macOS app.
  2. Open the downloaded .dmg file and drag Codex to your Applications folder.
  3. Launch the app and sign in with your ChatGPT account.
  4. Verify the app is updated to support the latest features (background computer use, plugins).

Step 2: Install and Configure the Computer Use Plugin

  1. In the Codex app, go to Settings (usually via the gear icon or sidebar).
  2. Navigate to the Computer Use section.
  3. Click Install for the Computer Use plugin.
  4. Follow on-screen prompts.

macOS will request permissions:

  • Screen Recording: Allows Codex to view screen content and take screenshots.
  • Accessibility: Allows Codex to click, type, navigate menus, and interact with UI elements.

Grant these in System Settings > Privacy & Security when prompted. Restart Codex if needed.

Step 3: Understand Computer Use Basics

Computer Use lets Codex:

  • See your screen via screenshots
  • Control the mouse (click, move, scroll)
  • Type on keyboard
  • Interact with any macOS app's GUI
  • Work in the background with its own cursor (doesn't interfere with your work)
  • Run multiple agents in parallel

It excels for tasks involving desktop apps without APIs/plugins, browser sessions, or multi-app workflows.

Key limitations:

  • Cannot automate Codex itself or Terminal in ways that bypass security
  • Cannot approve system security prompts
  • Best for scoped, narrow tasks

Step 4: Start Your First Computer Use Task

  1. Open a new thread in the Codex app.
  2. Prefix your prompt with @Computer Use or mention a specific app (e.g., @Chrome, @Notes).
  3. Describe the task clearly, including desired outcome.

Example prompts:

@Computer Use Play some focus music in Spotify and minimize the window.
@Computer Use Open @Messages, find the thread from Brooke about Yosemite, extract the best trip ideas, add them to a new note in Notes called "Yosemite ideas", then draft a reply.
@Computer Use Open Chrome, go to my local dev server, verify the checkout flow still works after changes.
  1. Approve any app access prompts (choose Always allow for trusted apps).
  2. Let Codex work in the background while you continue other tasks.

Codex will show progress, screenshots of actions, and ask for clarification if needed.

Step 5: Advanced Usage and Best Practices

Multi-App Workflows

Use natural language for complex flows:

@Computer Use Take interview notes from Notes app, add key points to Ashby tracker, update Slack status, and draft an email summary.

Browser Control

Specify the browser to avoid interfering with your session:

@Computer Use Use Chrome (not Safari) to test the new frontend at localhost:3000.

Background Parallel Agents

Run multiple independent tasks simultaneously without conflict.

Customization & Memory

  • Set defaults in customization (e.g., preferred browser).
  • Use memory features for recurring tasks.
  • Convert successful workflows into automations.

Safety Tips

  • Keep tasks narrow and scoped.
  • Stay present for sensitive actions (payments, credentials).
  • Review permissions carefully.
  • Close sensitive apps not needed for the task.

Common Issues & Troubleshooting

  • No screen visibility: Re-check Screen Recording permission in System Settings for Codex.
  • Cannot click/type: Verify Accessibility permission; restart Codex and macOS if stuck.
  • App not responding: Avoid running multiple Computer Use tasks on the same app simultaneously.
  • Mac locks/sleeps: Activity stops; keep machine awake.
  • Permission prompts loop: Use Always allow for trusted apps in Codex Computer Use settings.
  • Slow performance: Use for GUI-only tasks; prefer plugins/CLI for structured work.
  • Feature unavailable: Confirm subscription level and region eligibility.

For persistent issues, check Codex Troubleshooting section or restart the app.

Next Steps

  • Experiment with multi-step workflows across your daily apps.
  • Combine Computer Use with Codex plugins, in-app browser, and automations.
  • Turn successful tasks into reusable automations or skills.
  • Explore image generation and memory features for enhanced productivity.
  • Monitor OpenAI updates for Windows/Linux support and API expansions.

Practice with simple tasks first to build intuition for effective prompting.

Share this article

Referenced Tools

Browse entries that are adjacent to the topics covered in this article.

Explore directory
Codex mcp logo

Codex mcp

MCP Servers

Codex MCP Server is the MCP server mode of OpenAI Codex CLI. It lets MCP clients run and continue Codex coding sessions through a local stdio server.

openaiDeveloperDocs logo

openaiDeveloperDocs

MCP Servers

OpenAI Developer Docs MCP is an OpenAI-hosted, read-only server that lets MCP clients search and fetch documentation from developers.openai.com, platform.openai.com, and learn.chatgpt.com. It uses a public streamable HTTP endpoint and does not call the OpenAI API on the user's behalf.

ChatGPT Apps SDK logo

ChatGPT Apps SDK

Clients & Apps

The official OpenAI framework for building interactive apps that run inside ChatGPT. Built on the Model Context Protocol (MCP), it lets developers expose tools, structured data, and rich UI components directly within ChatGPT conversations.

code base memory mcp logo

code base memory mcp

MCP Servers

Codebase Memory is an open-source MCP server that indexes local codebases into a persistent knowledge graph for AI coding agents. It exposes structural code search, call-path tracing, impact analysis, architecture summaries, Cypher-style graph queries, ADR management, and related code-intelligence tools over stdio.

RevenueCat MCP Server logo

RevenueCat MCP Server

MCP Servers

RevenueCat MCP Server lets AI assistants manage RevenueCat subscription infrastructure through natural language, including projects, apps, products, entitlements, offerings, paywalls, and analytics.

Crunchbase MCP logo

Crunchbase MCP

MCP Servers

Crunchbase MCP is Crunchbase's official first-party remote MCP server for live company, funding, investor, people, acquisition, and predictive private-market intelligence. Interactive access uses a Crunchbase account with an assigned MCP seat and OAuth 2.1 authentication.