Is MCP here to stay, or just a stop on the way?

Every few months something in AI gets called "the standard", and a few months later it's old news. So it's fair to ask whether MCP will last.

On the numbers, it looks solid. It started at Anthropic, now sits under the Linux Foundation, and OpenAI, Google and Microsoft all support it. SDK downloads run into the tens of millions a month. That's not a fad. That's plumbing.

And as plumbing, it's brilliant. I'm building Latent Ink, the blog you're reading this on, MCP-first. I write, edit, tag and publish through Claude. I rarely open an admin screen. For getting things done, it's the best workflow I've had.

But here's the problem. We're visual people. Nobody falls for a blog because of its tool calls. People connect with how something looks and feels: the layout, the type, the images, the space around the words. A wall of text in a chat window isn't an experience. It's a transcript.

That's the bit Latent Ink is really testing. Can you build something people connect with when the main way in is a protocol? Right now I don't know yet. That's why I'm testing it.

The interesting thing is that the people behind MCP seem to agree. Its first official extension, MCP Apps, lets tools return interactive UI right inside the conversation: dashboards, forms, visualisations. Claude, ChatGPT and VS Code already support it. That's the protocol admitting text isn't enough.

So will MCP get left behind? I don't think so. I think it gets buried. It becomes the layer underneath that nobody talks about, like HTTP. What wins is whatever puts a good visual experience on top of it.

The question I'm left with isn't "MCP or not". It's where the experience lives. In the chat, in the browser, or somewhere we haven't built yet.

That's what I'm going to keep testing.