Codex Usage Guide

The recommended flow is to sign in with your own Codex account first, so Codex keeps its plugins, MCP, and Agent capabilities available. Then configure Magick API token mode to route model requests through Magick API. The Codex CLI and desktop app share the same local configuration.

1. Install And Sign In To Codex CLI

If Codex CLI is not installed yet, install it first:

npm i -g @openai/codex

Start Codex:

codex

The first run will guide you through Codex account authorization. Complete this step before configuring Magick API token mode so plugins and local tool capabilities remain available.

2. Configure Magick API Token Mode

Codex supports OpenAI-compatible protocols, so you can point model requests to Magick API with environment variables.

First get your Magick API key from the API Key Management page.

macOS/Linux

Apply only to the current terminal session:

export OPENAI_API_KEY="sk-your magickapi key"
export OPENAI_BASE_URL="https://api.magickapi.com/v1"

For a persistent setup, add the variables to ~/.zshrc or ~/.bashrc:

echo 'export OPENAI_API_KEY="sk-your magickapi key"' >> ~/.zshrc
echo 'export OPENAI_BASE_URL="https://api.magickapi.com/v1"' >> ~/.zshrc
source ~/.zshrc

If you use Bash, replace ~/.zshrc with ~/.bashrc.

Windows PowerShell

setx OPENAI_API_KEY "sk-your magickapi key"
setx OPENAI_BASE_URL "https://api.magickapi.com/v1"

After running setx, reopen your terminal before starting Codex.

3. Keep Codex Plugins Available

When using an API key or third-party OpenAI-compatible base URL, some Codex versions may hide or disable certain plugin capabilities, such as GitHub, Chrome control, Browser tools, or Agent-related tools. You can explicitly enable plugins in the Codex configuration file.

Configuration file path:

• macOS/Linux: ~/.codex/config.toml
• Windows: %userprofile%\.codex\config.toml

Append the following content to the end of the file:

[plugins."github@openai-curated"]
enabled = true

[plugins."chrome@openai-bundled"]
enabled = true

[plugins."browser@openai-bundled"]
enabled = true

If you use other plugins, continue enabling them one by one by plugin name. If your current Codex version supports a global plugin switch, you can also add:

[plugins]
enabled = true

If the configuration file already contains MCP or other plugin settings, keep the existing content and append only the enable entries above.

4. Restart And Verify

After saving the configuration, restart Codex:

codex restart

You can also exit the current Codex session and run:

codex

In Codex interactive mode, enter:

/plugins

Confirm that GitHub, Chrome, Browser, MCP, or Agent-related plugins are still available. When you call GPT models, Codex reads OPENAI_API_KEY and OPENAI_BASE_URL and sends requests to Magick API through https://api.magickapi.com/v1.

Alternative: Write The Codex Provider Configuration

If you prefer not to rely on environment variables, you can write Magick API into Codex's provider configuration.

Configuration file path:

• macOS/Linux: ~/.codex/config.toml
• Windows: %userprofile%\.codex\config.toml

model_provider = "OpenAI"
model = "gpt-5.5"
review_model = "gpt-5.5"
model_reasoning_effort = "xhigh"
disable_response_storage = true
network_access = "enabled"
windows_wsl_setup_acknowledged = true
model_context_window = 1000000
model_auto_compact_token_limit = 900000

[model_providers.OpenAI]
name = "OpenAI"
base_url = "https://api.magickapi.com/v1"
wire_api = "responses"
requires_openai_auth = true

Then write the API key into the auth file.

Auth file path:

• macOS/Linux: ~/.codex/auth.json
• Windows: %userprofile%\.codex\auth.json

{
  "OPENAI_API_KEY": "Key generated in the backend"
}

Notes

• For OPENAI_API_KEY, fill in the key generated in the Magick API dashboard.
• OPENAI_BASE_URL or provider base_url must be set to https://api.magickapi.com/v1.
• Token mode only routes model requests to Magick API. Codex account authorization keeps Codex's own capabilities, plugins, and tool experience available.
• In this document, both model and review_model use gpt-5.5. You can switch to other models supported in the [Model Plaza] as needed.