From a Quick Spec to an MCP-Enabled Blog

A blog site is a simple idea. But having a place to jot down notes and articles that might be useful to someone else is valuable. I’d been meaning to make one for a while, and being able to start with an AI prompt made the idea accessible enough that I finally stopped putting it off.

I wanted a straightforward site where I could collect ideas and articles about things I work with regularly. I wrote a build spec, then used an AI build agent to turn it into a working application. With a clear scope and my systems architecture background, I went from concept to launch in about an hour.

The spec helped keep the project grounded. I wanted a public site for articles, an admin area for writing and publishing, and a way for an AI client to work with the content. The result uses Go and SQLite, with categories for different topics and a small set of publishing tools.

That last part is MCP, the Model Context Protocol. It gives an AI client a defined way to interact with an application. For this blog, the MCP tools can list categories and posts, create an article, and publish it. The calls go through the site’s normal validation and content-sanitizing path, so publishing through an AI prompt still uses the application’s own rules.

That made the difference between having a blog and having a blog that fits into how I already work. I can ask for help drafting an article, review the copy, choose a category, and publish through MCP. The work feels native to the conversation instead of becoming another task I have to remember to do later.

Getting the application onto my VPS brought its own practical details. The server already runs nginx for several sites, so I configured nginx to keep serving them while routing lwi3.net to the new app. I started with a blank database, checked that the public site and admin sign-in worked, and verified that the other sites were still responding.

The first posts made the workflow real. I used MCP to publish an article about PracticeEngine, my music-practice project, under Music. I also used it to publish other articles through the same interface. The useful part is not simply that an AI can write text; it can work with the application in a defined way, using the category and publishing functions the site provides.

My takeaway is simple: when you build an application, think about making it accessible to MCP early. Give AI clients clear tools for interaction and access to the data that makes sense for the app. Telemetry and other capabilities can be part of that design as the application grows. A small, deliberate interface can make the system easier to work with, and can put those capabilities within reach of a prompt.

The blog itself is modest. That’s the point. I had ideas I wanted to share, and the combination of a quick spec, a capable build agent, and MCP made it easy to finally put them somewhere useful.