How I Use AI Agents + MCP to Fully Automate My Website’s Content

Source: Dev.to

Before AI

First of all, let me tell you what it is like to add a new podcast episode to my site. I use Nuxt Content, so each podcast is just a markdown file with some YAML front‑matter. This YAML contains things like the date, the name of the podcast, the image URL, the host, etc.

Last year I was simply getting an old podcast episode, clicking Duplicate in VS Code, and then renaming everything with the new podcast information. In other words, I manually:

1. Clicked the link to the new episode.
2. Copied and pasted the information from the hosting site into my markdown file.
3. Downloaded the episode image, uploaded it to Cloudinary, copied the Cloudinary image name, and pasted that into the file.

Cloudinary is great for managing images and keeping my site performant, but the extra work of downloading and uploading images was tedious. Sometimes it took me ages to add new episodes because I simply couldn’t be bothered.

Automating with Prompts and the Playwright MCP

I started to automate part of the process by using reusable prompts in VS Code. I created instructions for Copilot, then pressed the Play button to run the prompt in a new chat and gave it the podcast URL.

With the Playwright MCP installed, Copilot could navigate to the URL where the podcast was hosted and extract the relevant metadata. It worked well, saved a lot of time, and even allowed bulk updates by supplying multiple URLs.

The image handling was still an issue, though. I was tempted to stop using Cloudinary and store images directly in the public folder of my site, but that would mean losing Cloudinary’s optimization benefits.

Automating with Goose, Playwright MCP, Cloudinary MCP & GitHub MCP

Next I tried Goose, a coding agent from Block. Goose is a desktop app (with a CLI option) that lets you run “recipes” – essentially prompts with parameters. I copied my existing prompt into Goose and saved it as a recipe, adding the podcast URL as a parameter.

What the new workflow looks like

1. Playwright MCP navigates to the podcast page and extracts all needed information, including the image URL.
2. Goose downloads the image locally.
3. The Cloudinary MCP uploads the image to my Cloudinary account using the API key stored in the extension’s settings, automatically choosing the correct folder.
4. Goose writes the markdown file with the correct Cloudinary image ID.
5. Using the GitHub MCP, Goose creates a pull request with the new episode files.

I could even ask Goose to run the dev server and verify the changes with Playwright, taking a screenshot of the podcasts page to confirm everything looks right.

Everything just worked. After the PR was merged, the new podcast episodes appeared on my site with properly optimized Cloudinary images.

Conclusion

It took time to figure everything out and set up the process, and yes, copying and pasting would have been quicker for a single episode. But now the workflow is automated: the next time I want to add a podcast episode, I simply run my Goose recipe and pass in the podcast URL. The whole pipeline—from metadata extraction to image upload and PR creation—happens automatically.

I am a guest on one tonight so when that is out I will be able to add it easily to the site. In fact, if I had a team of people working on my site I could even share the recipe with them and they could simply run it.

I am using **Nuxt Content** for my site, which means I have no CMS. My content lives in markdown files and it makes it very easy as a developer to add content but perhaps not so easy for non‑developers. But now, even my mother could add a new podcast episode to my site. That is just amazing. This is just my personal site but think about the possibilities of this use case for many other businesses.

I am very impressed with what **Goose** can do. The more I am using it, the more it is blowing my mind. I am now going to go ahead and add other **recipes** for the rest of the content I add, or perhaps just modify this recipe with parameters so I can have one recipe. I shall keep playing around. This is fun.

Let me know if you found this interesting, are doing something similar, or have used any of the MCPs mentioned above. We are living in exciting times, so if you haven't started to experiment yet then what are you waiting for? Just play around and have fun.