I haven't opened my blog editor in months. I publish by talking to Claude.

I haven't written a post the old way in months. No opening the editor, no fighting with MDX, no manually uploading images. I tell Claude what I want to say and it publishes. And the most important part: what I do with my blog can be done with almost any internal business process.

The problem: publishing had too much friction

Running a blog for a small company sounds easy until you try to keep it going. My workflow used to look like this: open the editor, write in MDX, remember the component syntax, upload images to a bucket, copy URLs, validate nothing breaks, publish. Every post was two hours of friction for 30 minutes of thinking.

Predictable result: I published less than I wanted.

I've spent years building custom software for other companies with exactly this pattern: identify the friction, remove it. It was time to apply it to myself.

What is MCP, without the jargon?

Before going further, one technical term: MCP. It stands for Model Context Protocol and it's an open standard created by Anthropic. The most honest analogy is this: MCP is the USB port of AI.

Before, if you wanted an assistant like Claude to access your CRM, your database, or your calendar, every connection was a different cable, a different workaround. Every company invented its own way to plug tools into AI. Chaos.

MCP standardises that plug. Any tool that speaks MCP connects to any AI that speaks MCP. Full stop.

What I built: an MCP for my blog

I built a custom MCP to manage the Evolve2Digital blog. I gave it nine tools, each with a clear purpose. Some read, some write, some upload images, some validate that a post is correct before publishing.

The nine tools are grouped by risk. What the AI can do without asking for permission, and what it can't.

That separation isn't aesthetic. It's the difference between AI being useful and AI being dangerous. An AI that can delete content without asking is a problem waiting to happen.

What the real workflow looks like

A typical conversation between me and the MCP goes like this:

1. I tell Claude: "write a post about X, in this tone, with this structure"
2. It proposes an outline. I refine it.
3. It drafts section by section using the blog's visual components (callouts, pull quotes, lists).
4. If it needs an image that doesn't exist, it asks me to upload it. I upload. It references it.
5. Before publishing, it automatically validates everything is correct: images exist, frontmatter is right, no broken links.
6. Only then does it publish.

The key point: I decide at every step. The AI proposes, I approve. No surprises.

“

The AI doesn't replace me as a writer. It removes the boring part of the craft: formatting, images, validation. What adds value — deciding what to say and how — stays mine.

”

This isn't about blogs

If you're reading this and you run a small business, you probably don't care much about blogs. Fair. The blog isn't the point. The pattern is.

The same approach works for almost any internal system with friction:

In every case the recipe is the same: identify the process with friction, define what tools the AI needs to execute it, set clear limits (what requires confirmation, what doesn't), and connect everything via MCP.

What used to be "let me open five programs" becomes "I'll just tell Claude".

What this actually unlocks

There's a detail most people miss: when you reduce the friction of a process to near zero, you do it more often. It's not just efficiency. It's behaviour.

I publish more because it costs less. A business that automates quote creation doesn't just make them faster: it makes more quotes, chases more opportunities, and sells more. Friction doesn't just steal your time. It steals actions you never end up taking.

The detail that matters

What I'm describing isn't theory. It's the system I use every day. And the recipe to build it isn't just for blogs: any repetitive internal process, with clear rules and operational friction, is a candidate for something like this.

The useful question isn't "can this be done in my company?". Almost always it can. The useful question is "which process is costing me more time than it looks like?". That's where the ROI starts.