Since the introduction of Model Context Protocol (MCP), I thought this was just gonna be another hype extension that’s simply just a Chatgpt wrapper. I started playing with the git projects my tech team downloaded and it was cool. They described it as “toybox for developers”
I don’t even know what that means.
I had to ask chatgpt what MCP was. Went down this rabbit hole and the answer is “ MCP is about giving AI models structured, contextual access to tools, data, and workflows without hardcoding every single integration. Because one of the most frustrating parts about coding AGI’s is the constant reminders and making sure educating prompts don’t conflict with each other. So far it seems like that problem has been solved with what I’m guessing is like an API layer for context.”
Ok cool. So MCP lets me connect models to specific sources of data, whether that’s our internal CRM, my calendar, or some obscure project database. But instead of a single query, it can query, act, and respond within a single request in multiple areas.
Let’s say I’m running a project for a supply chain operation. Normally, if I ask a model, “What’s our average lead time for the past 3 months?” it will either:
Hallucinate, or
Ask me to go pull the data.
With MCP, I can configure a tool that bridges the AI to the actual warehouse database or ERP dashboard. Then I can write a request like this:
{
"tool": "supply_chain_db",
"action": "query",
"params": {
"metric": "lead_time",
"timeframe": "last_90_days"
}
}
The model doesn’t just guess. It will call the tool, pulls the data, and replies with something like:
“Your average lead time over the last 90 days is 17.2 days, down 1.8 days from the previous quarter.”
Actually maybe that’s not a good example to use because I just made it sound like something a function-call or API integration could’ve done. Think of it like MCP knows to talk to multiple systems before giving me the answer instead of the answer coming from one simple code.
Ok wait, instead of just asking for warehouse lead times, I could ask: “What’s our on-time delivery performance this quarter, and can you draft a summary email to the head of ops with the numbers and next steps?”
This way, MCP will perform the query to our ERP for delivery data, cross-reference it with CRM sales forecasts, format the insights into a message, and then push it directly into my outlook or somewhere else all within the same interaction. That’s what makes this cool: it turns isolated integrations into a reusable, context-aware ecosystem, letting AI act more like a true operator’s assistant rather than a one-off script runner.
It’s cool. I don’t know if it’s groundbreaking but pretty cool first step and nice to have but not a need. I can just describe tools and let the model use them when relevant. My developers like it because there’s no need for plugins either with the model’s reasoning loop. They want to get a MCP setup to use it across several areas of their work. Not sure if I’m ok with that but I really can’t stop them. Pretty cool either way.
Not my turn for permanence
February 8, 2025