The MCP Cat Is Out Of The Bag

Toolmakers and service providers are you listening?

(Note: This is a follow up to my article about Anthropic’s recent Claude MCP announcement.)

Let’s start with the obvious: I use ChatGPT to help draft my articles. Including this one.

Not just a little “AI as thesaurus” help—I mean real scaffolding, restructuring, helping me find the center of the idea. That’s not a secret. What’s changed is this: the workflow around that help is now the bottleneck.

What’s Actually Broken

After I finish working with GPT—usually in chunks, iteratively—the annoying part kicks in. I copy everything over to Substack. Then I spend another 10–15 minutes reformatting the structure, updating metadata, previewing links, doing the fiddly UI dance. It’s not fatal. But it’s friction. And for longer pieces, it breaks flow.

The real issue isn’t copy-paste fatigue. It’s separation of concerns—between where the work happens and where it ends up. That’s the piece that feels increasingly out of date.

The Chunked Generation Advantage

GPT is stronger in iteration than monologue. Long pieces benefit from chunk-by-chunk feedback. But that requires you to maintain state: what was said, what got revised, what changed. Right now, I track that manually in Notion, memory, or markdown snippets floating around my desktop. It’s fine. But the moment the source lives in the destination—when the AI works within Substack or any editor like it—that whole process simplifies.

You don’t need to restitch your own thoughts just to hit “Publish.” (Or schedule or add tags. Or teaser images. Or add a coordinated Taplio post. You get the idea.)

This Isn’t Just About Convenience

Sure, this could be seen as a “writer’s workflow” gripe. But zoom out. This is a microcosm of what’s coming for every productivity tool. As LLM-native behaviors become the norm, we’ll expect our tools to stop being static containers and start acting like shared workspaces—with AI as a first-class participant.

The pressure on both AI providers (OpenAI, Anthropic, etc.) and tool platforms (Substack, Notion, Figma, Google Docs) is only going to grow. It’s not just about integrations. It’s about embedding capability—so users don’t have to bridge the gap themselves.

That pressure doesn’t come from enterprise checklists or vendor RFPs. It comes from people like us—tinkering, drafting, iterating—and wondering why the smartest part of the process still ends in manual copy-paste.

So, Who Wins?

We do.

We get tools that match how we actually work. We get workflows where AI supports flow rather than interrupting it. And we get back a little more of what makes writing—and building—fun: the part where ideas move fast and stay connected.

The cat’s out of the bag. Now let’s build the tools that know it.

Afterthought:

If I were Zapier, Make, Retool, or any provider of low/no-code tools, I’d be furiously working on pivoting to become an MCP integrator or registry for internal apps and custom workflows. That long tail of use cases won’t get direct model support anytime soon—and their current market is about to get squeezed by AI-native capabilities.