AI Tools Are Finally Learning to Talk. That's Not the Same as Being Connected.

Every few months a new number makes the rounds in AI. This summer it's the count of MCP servers. The Model Context Protocol — the open standard that lets AI models talk to outside tools — now counts thousands of these connectors, and by mid-2026 every major vendor supports it. Anthropic, OpenAI, Google, Microsoft. The protocol shipped a fresh specification on July 28, 2026, and the ecosystem keeps growing.

The number is real. What it means for your business is more complicated, and mostly better than the headline suggests.

What actually changed for us

For years, connecting one tool to another meant writing a bridge by hand. A client uses a CRM, an accounting app, and a messaging channel. None of them speak the same language. So we wrote custom code for each pair — read from here, translate, write to there. Every new tool meant another bridge. Every API change meant fixing one.

MCP cuts a lot of that work. When a tool ships an MCP connector, an AI agent can use it directly. We don't write the plumbing; we write the logic that decides what to do with it. That is the part of the job that was always worth paying for. The rest was overhead.

So the honest version of the news is this: the standard is doing what standards do. It removes repeated work. It doesn't remove the thinking.

Why the big number is the wrong thing to watch

Thousands of servers sounds like everything is ready. It isn't.

Most of those connectors are for developer tools and popular platforms. The specific stack a small business runs every day — the local accounting software, the industry booking system, the CRM your team actually uses — is often not on the list. And when a connector does exist, quality varies. Some are maintained. Some were published once and abandoned. A big count on a directory tells you the ecosystem is active. It doesn't tell you your tools are covered.

This is the part we push back on. A standard being popular is not the same as your problem being solved. Someone still has to check whether your tools are in the ecosystem, decide what the agent is allowed to do, set the permissions, and handle the cases where something fails. The protocol makes the connection possible. It doesn't make the decisions.

What we actually use it for

We're not describing this from the outside. MCP is already part of how we work.

We use the Plane MCP for planning. It helps us lay out the work, and — this is the part clients notice — it opens accounts for them on top of Plane where they see exactly what's being done. The tasks are visible. The status is honest. For us, this MCP plays a real role in handing work to clients transparently, instead of asking them to trust a status update in an email.

We recently started using the HiggsField MCP for creative work. Custom design, video, image generation — the jobs that need something made, not just moved. Having that inside our workflow instead of as a separate tool is a clear advantage.

We also use the HeyGen MCP for social media content. It's a real help in managing content, and clients are genuinely happy with the avatar work — putting a consistent identity on screen without a full production every time.

Three tools, three different jobs: planning, creative production, content. None of them required us to build a bridge from scratch. That is the practical version of what the big number is pointing at. Not "everything is connected now," but "the connections we needed were there, and we spent our time on the work instead of the wiring."

The counterpoint that keeps us honest

None of this means you should rush. The same standard that saved us the plumbing can tempt a business into thinking connection equals value. It doesn't.

An agent that can reach your accounting system is powerful and risky in equal measure. The question is never "can it connect." The question is "what should it be allowed to do, and what happens when it's wrong." Those answers don't come from the protocol. They come from someone who understands your business and takes responsibility for the setup. This is the same reason we keep a human in the loop on anything that touches money or a customer directly.

So we read the 10,000-server headline the way we'd read any inventory count. Useful signal. Not a finish line.

What to do tomorrow

If you run a business and you want one concrete step, here it is: make an inventory of your tools.

List every system your team touches in a week. The CRM, the accounting app, the booking tool, the messaging channels, the spreadsheet nobody admits to. Then draw the lines that don't exist — the places where someone copies data from one screen into another by hand. That copying is where the time goes. That list is your integration map.

You don't need to know anything about MCP to make it. But once you have it, you'll know exactly which connections are worth building, and you'll be able to tell quickly whether the standard already covers them or whether they need real work. That is a far better starting point than a directory of thousands of connectors, most of which have nothing to do with your business.

The tools are finally starting to talk to each other. That's genuinely good news. It just doesn't replace the decision about what they should say. If you want help drawing that map and connecting the pieces that matter, that's the core of our workflow automation work.