What on Earth is MCP? (And Why We Won't Shut Up About It)

If you've spent any time around us lately, you've probably heard the letters "MCP" thrown around like we've joined a cult. We haven't. Well — mostly haven't. Let's explain what it actually is, why we're borderline obsessed with it, and how it's quietly powering a lot of the automation behind New Forest Technology Services.

So, what actually is MCP?

MCP stands for Model Context Protocol. Think of it as a universal socket adaptor for AI. Normally, if you want an AI assistant to actually do something useful — check your calendar, read an email, look at your invoicing system, control a smart building — someone has to build a bespoke, fiddly, one-off connection for every single tool. It's the digital equivalent of every country having a different plug socket.

MCP is the international travel adaptor. It's an open standard that lets an AI model talk to external tools, apps and data sources in a consistent way. Build one MCP "server" for a tool, and any MCP-compatible AI can use it. No more bespoke wiring for every combination.

Why we're genuinely excited about it (not just marketing fluff)

• One protocol, endless tools — instead of ten custom integrations, you build to one standard and it just works with any compatible AI.

• It turns "AI chatbot" into "AI that can actually do things" — check real data, trigger real actions, control real systems.

• Massive time saver for automation-heavy setups like ours — n8n, Vultr servers, Xero, Google Drive, KNX systems — all become things an AI can be pointed at directly.

• It's open and growing fast, so the tooling ecosystem gets better every week without us having to build everything ourselves.

• It keeps sensitive access controlled — you choose exactly which tools and data an AI can see, rather than giving it the keys to everything.

The Do's

• DO start with read-only tools first (search, list, view) before you let anything write, delete or send.

• DO scope permissions tightly — an MCP server should only be able to do what it actually needs to do, nothing more.

• DO test in a sandbox or staging environment before pointing an MCP server at anything that touches real customer data or live systems.

• DO keep an audit trail — know what the AI did, when, and why.

• DO document your MCP servers like you would any other piece of infrastructure. Future-you will thank present-you.

The Don'ts

• DON'T give an MCP server admin-level access "just in case." That's how you end up explaining to a client why their entire database got creatively reorganised.

• DON'T connect an AI directly to production systems without a human approval step for anything destructive (deleting, sending, paying, publishing).

• DON'T assume every MCP server you find online is safe — treat third-party servers with the same scepticism you'd give a random USB stick.

• DON'T skip the error handling. AI models are confident even when they're wrong; your tools shouldn't be.

• DON'T forget that "it worked once" is not the same as "it's ready for production."

The best way to think about MCP: it's not magic, it's plumbing. Good plumbing, done properly, is the difference between a smart building and a very expensive puddle.

Where this fits into what we do

Behind the scenes at NFTS, MCP is already doing quiet, unglamorous, very useful work — connecting our automation stack, our invoicing, our infrastructure monitoring and more into tools we can actually talk to directly. It's less "robot uprising" and more "finally, the boring admin bits sort themselves out."

If you're a business owner wondering whether any of this applies to you: if you've got systems that don't talk to each other, or you're doing the same manual task for the hundredth time, there's a good chance MCP-powered automation could take it off your plate.

---

Fancy finding out what MCP could automate away in your business? Get in touch with New Forest Technology Services and we'll have a chat — no jargon required, promise.