"Agents" and "MCPs": the plain-English version

You might have seen "agents" and "MCPs" pop up in the AI news, the Twitter posts, or the wandering opinions of someone in your family who reads Hacker News. Both terms mean something specific. Both are early. And both are pointing at where AI is actually heading next.

Agents, in one sentence

An agent is an AI that takes actions. Where a normal chatbot describes what you should do, an agent goes and does it.

A regular ChatGPT conversation looks like this: you ask a question, you get an answer, and you go off and act on the answer yourself. An agent skips that last step. You tell it "follow up with this customer if they haven't replied by Friday," and the agent monitors your inbox, drafts the message in your voice, and sends it on the day.

Right now, agents are early. They get things wrong. They get confused in long workflows. They sometimes do exactly what you said instead of what you meant. None of this is going to be true in two years, but for this year, the safer play is to use them for low-stakes work first. Things where a mistake costs you a few minutes, not a customer.

Good first agent tasks for a small business:

• Tagging or sorting incoming email
• Drafting (not sending) follow-ups
• Reformatting messy spreadsheet data
• Watching a folder for new files and renaming them

The agent picks up on patterns you've already established and does the boring repetition. Most owners we work with start there and slowly hand over more.

MCPs, in one sentence

MCP stands for Model Context Protocol. It's a standard way for AI tools to plug into your other tools.

If you've ever connected Zapier to your apps, that's the rough idea, except instead of you writing rules ("when a new lead comes in, do X"), the AI tool just gets a live connection to your tools and figures out what to do based on what you ask.

The reason MCPs matter: until recently, ChatGPT and Claude couldn't actually see your Google Drive, your CRM, or your invoicing software. You had to copy and paste. With MCPs, the AI can read and write directly. So when you ask "what's the status of the Anderson project," the AI looks it up in your actual project tracker, instead of you pasting the dashboard.

That's the practical shift: AI that used to live in a separate browser tab is moving inside your business.

What to actually do about all this

If you're hearing about agents and MCPs and feeling behind, you're not. Most of this has been usable for less than a year. The companies that have figured it out have a handful of internal experiments running, not a polished system.

What matters more than understanding the terms is having clean, well-organized context for your business. Without that, agents do dumb things and MCPs connect to a mess. With that, both become genuinely useful.

If you want to read more about why context is the piece everything else depends on, start here.