Install WordPress on my account and set up a backup schedule
Type this in the Invoker Assistant and it doesn't matter whether you've connected Claude or Codex. Both carry out the same instruction on your hosting. The difference lies somewhere else.
Same tasks, two providers
In the Invoker Assistant, Claude Code and Codex do the same work:
- installing WordPress,
- editing a theme or plugin,
- connecting a repository and deploying over Git,
- making or restoring a backup,
- setting up mail and DNS,
- taking a screenshot of your site, on desktop or mobile.
Both work on your own hosting account, with the same notes, the same previews and the same access to your files. Neither gets root access or access to other accounts on the server.
What actually differs: the model and your subscription
The real difference comes down to two things:
- The model. Claude Code runs on Claude (Opus, Sonnet or Haiku). Codex runs on GPT-5.5, GPT-5.6 (Sol, Terra or Luna) or GPT-6-Astra. Each model has its own way of reasoning and writing.
- The subscription you pay for it. Claude Code uses your Claude subscription (Pro, Max, Team or Enterprise). Codex uses your ChatGPT subscription with Codex. Your hosting price doesn't change, whichever you pick.
Switching per conversation
You connect both side by side through Invoker Link, on the Claude and Codex tabs. In the Invoker Assistant you then pick the provider and model per new conversation. A conversation already underway keeps the model it started with; a new one starts with whatever you choose.
Approvals: almost the same, in a slightly different order
With Claude Code you decide how much you see beforehand: approve every change one by one, let file changes go through automatically, or full auto mode. Codex works a bit differently: changes go through directly and then show up afterwards as a card with the exact diff. Either way, you see what changed, and you can undo a whole task in one go.
Want to decide per change beforehand? Pick Claude Code. Want a task finished in one go so you can review the result afterwards? Codex works just as well for that.
| Claude Code | Codex | |
|---|---|---|
| Model | Opus, Sonnet or Haiku | GPT-5.5, GPT-5.6 (Sol, Terra, Luna), GPT-6-Astra |
| Subscription | Claude Pro, Max, Team or Enterprise | ChatGPT with Codex |
| Approvals | per change, automatic, or auto mode | change applied directly, then a card |
| Undo | per task | per task |
| Notes and previews | shared between both | shared between both |
| Scope on your hosting | your own account, or admin on a Claude Server | your own account, or admin on a Claude Server |
Our advice
- Use what you're already paying for. If you already have a Claude subscription or a ChatGPT subscription with Codex, start there. There's no reason to take out a second subscription just for your hosting.
- Have neither yet? Pick based on what you already know from other work. The experience in the Invoker Assistant is comparable for both.
- Not sure? Connect both and try the same instruction on a test subdomain with Claude Code and with Codex. It costs you nothing extra and you'll immediately see where the model makes a difference.
Common mistakes
Expecting Codex to ask for approval per change. It doesn't. It carries out the instruction and shows a card afterwards. Want to decide per step? Use Claude Code instead.
Taking out a subscription you don't need. Your hosting plan works the same, with or without Codex, at the same price. Only connect Codex once you already have a ChatGPT subscription with Codex, or plan to get one for other reasons.
Trying a big task straight on your live site. Test an unfamiliar model on a test subdomain first, especially for a task that could break your site, such as a major WordPress update.