What Is MCP, and Why Should You Care?
Founder, Automation Squad ·
MCP sounds like protocol jargon. It's actually the plug that lets your AI assistant reach your calendar, your CRM, your accounting software — instead of just talking about them. Here's what changed in 2024 and what to ask before you turn it on.
By the Automation Squad Research
Your AI assistant is brilliant at conversation and useless at your business, and the reason isn't the model. It's that the model can't see anything. Ask Claude or ChatGPT what your top customer ordered last month and it has no idea — not because it's not smart enough, but because nobody's ever shown it your order system. MCP is the plumbing that fixes that. Skip the acronym for a second and just remember what it does: it lets your AI actually reach the software you already run your business on.
That's the whole decision, really: know what MCP actually gives your AI access to, and ask the security question before you flip the switch — not after.
The problem nobody built a fix for until 2024
Before MCP, connecting an AI model to a business tool meant custom code. A developer had to hand-build a bridge between, say, ChatGPT and Salesforce. Then a different bridge for QuickBooks. Then another for Gmail. Every AI tool needed its own custom connector to every piece of software, and every piece of software needed its own custom connector to every AI tool. Multiply ten AI tools by fifty business systems and you get five hundred bespoke integrations, each one built and maintained separately. Nobody was going to build all of those. So most AI tools stayed stuck talking to themselves — brilliant at language, blind to your actual data.
What MCP actually is, minus the jargon
Anthropic released the Model Context Protocol as an open standard in November 2024, and described it with an analogy that's held up well: think of it as a USB-C port for AI. Before USB-C, every device had its own charger and its own cable. USB-C gave everything one shared plug. MCP does the same job for software — instead of a custom-built bridge between every AI and every business tool, a company builds one MCP "server" for their software once, and any AI that speaks MCP can plug into it. Your AI doesn't need to know the inner workings of QuickBooks or your calendar. It just needs the plug.
Who's actually using it, not just talking about it
This isn't a niche Anthropic project. OpenAI added support for MCP across its tools in March 2025 — a notable move, given OpenAI and Anthropic compete directly. Google DeepMind confirmed support for MCP in its Gemini models the following month. Microsoft built MCP support into Copilot Studio and its Windows AI platform. When your three biggest AI vendors all adopt the same connector standard within months of each other, that's not a fad — that's the plumbing setting, the way HTTP set the plumbing for the web. Software companies have been shipping their own MCP servers ever since: project management tools, CRMs, accounting platforms, calendars, databases. If a business tool has added AI features in the last year, there's a decent chance it now speaks MCP.
What this means for a task you'd actually assign
Say what you want done, not what MCP does in the abstract. "Pull this week's unpaid invoices from QuickBooks and draft follow-up emails" used to require either a developer building a custom integration or you doing it by hand. With an MCP connector to your accounting software switched on, an AI assistant can look at the real invoice data, draft the emails referencing actual amounts and due dates, and hand you a batch to review before sending. Same story for a calendar connector — "find a 45-minute slot next week that works for me and the three people on this email thread" only works if the AI can actually see your calendar, not guess at it. MCP is the difference between an AI that talks about your business and one that can look at it.
The question to ask before you turn any of this on
Reach, once granted, is reach. An MCP connector that lets your AI read your invoices can usually also let it act on your invoices, depending on how it's configured — and a connector built by a smaller vendor may not have had the same security scrutiny as the big platforms. Two questions settle whether a given connector is safe to enable: does it default to read-only, or can it take actions without asking? And can you scope it — grant access to, say, your calendar but not your entire email archive? If a vendor can't answer both clearly, that's your answer. Ask before you connect anything to systems holding customer data, financial records, or anything you'd hate to see leak or get modified by mistake.
You don't need to build this — you need to know to ask for it
None of this requires your business to write a line of code. MCP is infrastructure other companies build; your job is to know it exists so you can ask the right question when a vendor pitches you AI features. "Does this connect to our other systems, or does it only work inside your app?" is now a fair, sharp question to ask any software salesperson — and if MCP or an equivalent isn't part of the answer, you're likely buying an AI feature that stays walled off from everything else you run.
Two things to do this week
- Ask your top three software vendors — the CRM, the accounting platform, the project tool — one direct question: "Do you support MCP, or another way to connect your data to an AI assistant?" Write down the answers. That list tells you which of your existing tools are actually ready for AI to touch, and which aren't yet.
- Before turning on any connector, confirm in writing whether it's read-only by default and whether access can be scoped to specific data. Don't take "it's secure" as an answer — ask for the specific permission model.