MCP in plain English: how we connected AI to a client's CRM last week

Model Context Protocol. Four syllables that mean almost nothing if you haven't been deep in AI development circles. But what it does is something every small business owner should understand — because it fundamentally changes what AI can do for your operation.

Let me skip the technical definition and start with what happened when we used it last week.

What we actually did

A client runs a specialty contracting business. They use Zoho CRM to manage clients, projects, and follow-ups. Like most small business owners, they spend a significant portion of their day looking things up — pulling up client records, checking what stage a project is in, figuring out which follow-ups are overdue, generating status updates.

We connected their CRM to an AI assistant using MCP. Now, instead of navigating to CRM, running a search, finding the record, reading through it, and then manually creating a follow-up task — the conversation goes like this:

"What's the status on the Henderson project and are there any outstanding follow-ups?"

The AI looks up the record in real time, reads the relevant fields and notes, checks the task list, and responds with a plain-language summary — plus the option to create a follow-up task or draft a status email right there.

That's MCP. The AI isn't guessing. It's actually looking at live data in your actual CRM.

So what is MCP technically?

MCP stands for Model Context Protocol. It's an open standard that lets AI models connect to external tools and data sources in a structured, secure way. Before MCP, AI assistants were islands — you could ask them questions, but they couldn't see or touch anything in your actual business systems without a lot of custom integration work.

MCP changes that. It creates a standardized "handshake" between an AI model and an external system, so the AI can query data, take actions, and operate within defined boundaries — without you having to build custom integrations for every tool you use.

What makes it different from regular AI assistants?

A regular AI assistant answers questions based on what it was trained on and what you paste into the conversation. It's helpful. It's not connected to your business.

An MCP-enabled AI assistant can look up your CRM record, check your calendar, create a task in your project management tool, pull data from your accounting system, and send a draft email — all in one conversation. It's not answering questions. It's doing work.

What did the setup actually look like?

For this particular client, we set up an MCP server that connected their AI assistant to Zoho CRM. We defined exactly what the AI was allowed to do — read records, create tasks, update certain fields, and draft emails — and what it wasn't allowed to do (delete records, change financial data, send emails without human review). That last part is important: you're always in control of the scope.

Then we integrated it into the workflow they already had. Same tools, same process — but with AI handling the lookup-and-retrieve steps that used to eat time.

MCP is still early. The tooling is improving fast and the platforms supporting it keep growing. But it's real, it works, and we're already using it in production with clients. If this is the direction AI is heading — and it is — now is a good time to understand it.