MCP: The Quiet Standard That's Wiring AI Into the Tools You Already Run

Every few years a standard shows up that nobody outside the industry notices and everybody ends up depending on. USB did it for hardware. In 2026, the Model Context Protocol, MCP, is doing it for AI, and it's why your assistant can increasingly just read your CRM instead of you pasting things into a chat window.

MCP is an open standard, originally created by Anthropic and now governed by a foundation under the Linux Foundation with Anthropic, OpenAI, Block and others behind it. The pitch is exactly the USB analogy: one universal connector that lets any AI application plug into any data source or tool, without a custom integration built for every pair.

Why this stopped being a developer story

For its first year MCP was something engineers talked about. The numbers say it's now infrastructure.

When every major AI vendor and a growing share of business software agrees on one plug, the software you already pay for starts growing MCP endpoints whether you asked for them or not. HubSpot, Xero, monday.com, your job-management system, over the next year or two, most of your stack will be reachable by an AI assistant through this one mechanism.

What it changes for a business your size

Three practical shifts matter for SMBs.

1. The integration tax drops

A meaningful slice of what businesses pay for "AI integration" today is really custom plumbing: someone wiring model A to system B. MCP standardises the plumbing. The work shifts from can we connect it to should we, and with what permissions, which is a much better problem to have.

2. Your AI stops working from memory

An assistant connected over MCP answers "what's in the pipeline for September?" from the live portal, not from whatever you pasted into it last Tuesday. That's the difference between a party trick and a tool your ops manager actually uses on a Monday morning.

3. Access control becomes the whole game

Here's the flip side, and it's the part I'd underline. A protocol that lets AI reach into your systems is exactly as safe as the permissions on the connection. Security agencies have already published hardening guidance for MCP deployments, and the recurring themes are the unglamorous ones: authentication, audit trails, scoped access. The risk isn't exotic. It's an over-permissioned connector nobody remembers approving.

MCP makes connecting AI to your systems easy. It does nothing to make it wise. That part is still your job.

Four questions to ask before you plug anything in

1. What exactly can this connection read and write? "It connects to our CRM" is not an answer. Which objects, which fields, read or write? If the vendor can't say, don't connect it.
2. Whose credentials does it act under? A connection with admin rights is an admin. Give AI connections their own scoped identity so you can see, and revoke, exactly what they touch.
3. Where's the log? You want one place that shows every query and change the connection made this week. No log, no connection.
4. Does this replace glue, or judgement? MCP is brilliant at replacing brittle copy-paste and one-off integrations. It doesn't replace deterministic automation where you need guaranteed behaviour. Your invoice reconciliation should still be a workflow with defined inputs and outputs, not an AI improvising over a protocol.

Where this leaves n8n and your existing automation

A question we're getting a lot: does MCP make workflow automation redundant? No, they're layers. Deterministic workflows remain the right tool for anything that must happen the same way every time: billing runs, compliance documentation, settlement admin. MCP-connected AI sits alongside as the interrogation and drafting layer. The thing that reads across systems, answers questions and prepares work for a human or a workflow to execute. The businesses getting this right are composing the two, not choosing between them.

The takeaway

MCP is the kind of standard you don't need to understand deeply but do need to plan around: your stack is going to speak it, your AI tools already do, and the connections will be offered to you with a friendly one-click prompt. Say yes deliberately (scoped permissions, dedicated identities, a log you actually check) and it's the cheapest integration layer you've ever deployed. Say yes casually and you've opened doors you can't enumerate.

Deliberate is not much harder than casual. It's just earlier.