Updated: September 25, 2026

Using Codex CLI on macOS (Apple Silicon)#

Codex CLI is the terminal interface for OpenAI Codex.

Codex supports two authentication methods:

1. ChatGPT account

   • Uses ChatGPT plan limits (Free, Plus, Pro)
   • No API key required

2. OpenAI API key

   • Uses OpenAI Platform API billing
   • Requires an API key
   • Usage is charged against API credits

You can switch between both methods.

1. Installation#

Install Codex CLI via Homebrew Cask:

brew install --cask codex

Verify:

which codex
codex --version

2. Configuration#

Codex stores configuration in:

~/.codex/

Important files:

Edit configuration:

nano ~/.codex/config.toml

3. MCP Server Configuration#

MCP servers are managed through the centralized Git-backed pipeline documented in:

MCP Configuration Guide

Check MCP and system configuration:

codex doctor

4. Authentication#

Option A — ChatGPT Account Login#

Uses your ChatGPT account limits.

Login:

codex login

Check status:

codex login status

Example:

Logged in using ChatGPT

Check Codex usage:

/status

Online usage page:

https://chatgpt.com/codex/settings/usage

Option B — OpenAI API Key Login#

Uses OpenAI API billing.

Create an API key:

https://platform.openai.com/api-keys

Store the key in 1Password.

Add it to your shell environment:

nano ~/.zshrc

Add to ~/.zshrc:

alias codexa='printf "%s" "$OPENAI_API_KEY SEE 1Passoed" | codex login --with-api-key'

Then switch to API authentication:

codexa

5. Switching Authentication Methods#

Check current login:

codex login status

Switch to ChatGPT account:

codex login

Switch to API key:

codexa

6. Checking Usage#

Inside Codex:

/status

ChatGPT usage:

https://chatgpt.com/codex/settings/usage

API usage:

https://platform.openai.com/usage

7. Diagnostics#

Run:

codex doctor

Check configuration:

cat ~/.codex/config.toml

8. GPT-6 Missing from the ChatGPT Desktop App or Codex CLI#

If GPT-6 models appear in ChatGPT on the web but are missing from the local Codex model picker, check whether the local Codex configuration routes requests through Ollama or loads a custom model catalog.

In ~/.codex/config.toml, remove stale active settings such as:

openai_base_url = "http://127.0.0.1:11434/api/codex/v1"
model_catalog_json = "/path/to/an/ollama-model-catalog.json"

These settings can make the OpenAI model picker use the local Ollama endpoint and catalog. After removing them, restart the ChatGPT desktop app. For Codex CLI, restart its background app server so it reloads the configuration:

codex app-server daemon restart

This may disconnect active CLI sessions. Reopen the CLI and use /model (singular) to choose a model. To check the CLI’s bundled catalog without printing full model metadata:

codex debug models --bundled | jq -r '.models[].slug'

The separate ollama-launch profile intentionally selects an Ollama provider and custom catalog. Continue using it for local models; omit that profile when you want the standard OpenAI model catalog.

9. Quick Reference#

Notes#

Codex CLI uses the same installation for both authentication methods.

Use:

• ChatGPT login when you want to use included Codex access.
• API key login when you want API billing and usage tracking.