Sometimes learning looks like backpedaling.

And that’s okay.

I revisited an existing project (Letter Artdio) that had stalled, as my test case for Github connection within Codex and how to write natural language for a code update request.

The block plugin had initially been built at a time when I was not as clear as I am now, so it employed more ‘help’ from ChatGPT, while I used natural language to understand each step I would take, reading through the responses and code to grasp the concepts.

I spent two days diving into MCP to understand how I could use it in my WordPress development, and though it’s still a little murky, I’m at a far better place than yesterday.

I was able to achieve a few things:

• Set up MCP for my WordPress.com account
• Explored Codex CLI (still debugging the CLI side, but Codex is working inside ChatGPT)
• Connected Codex (ChatGPT) to my GitHub repo, requested a new feature via a branch, and tested changes
• *Used git apply inside Cursor’s terminal to patch the branch, but still figuring out full authorisation
• Refreshed about Interactivity API basics by triggering bounce animations on hover
• Began drafting a block-binding-powered homepage template for my custom post type – and then ditched block-binding, since that was overkill for what I actually wanted to do

*I still use Github Desktop at times to coach me through Git commands, and help me memorize them. It’s similar to how I treat Spanish learning, I think of the sentence in my head, write it out and only afterward I check whether I was ‘right’ or ‘wrong’.

All in all, though I was not able to do what I wanted with the plugin, it was a solid march forward on learning about OpenAI Codex, MCP connections, and a bit about block/plugin development.