MCP and LLMs: What it is and Why You Should Care as an Agribusiness Professional

MCP, Model Context Protocol, is an open protocol that standardizes how applications provide context to LLMs.

In Mental Models for Why the Future of Agronomy is Ai-Augmented not Automated I talked about two shortcomings of LLMs for agronomy specifically, but also why they more broadly had limitations:

1. Data Integration

2. Memory and Context Windows

MCP can alleviate the shortcomings in both of these areas.

An MCP is like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various accessories, MCPs provide a standardized way to connect AI models to different data sources and tools, improving data access and context. It's not just data stores like databases, either. MCP servers can expose various tools and resources to AI models, enabling functionalities such as querying databases, or interacting with messaging platforms like Slack or Discord, as basic examples today.

Evolution

We started with LLMs being basic chatbots with no ability to use any tools, or take any action. We could ask a question and get a response. Very basic.

Next, companies like Perplexity and ChatGPT built out tools and capabilities on top of the LLM enabling them to search, or take some specific action via a connection to a website. This is essentially an API, which works, but over the long term there becomes endless stacks of APIs that may not connect to other APIs, and a lot of edge cases that can fail.

That is where MCP come in.

Developers need standards. MCP is a standard that external developers can use to build connectivity from their systems to LLMs like ChatGPT or others. It allows external systems to communicate effectively, and seamlessly with LLMs delivering the capability to the LLM to access information and take actions. Data integration and context.

As a future example, if an ag retailer wanted a drone scouting solution that identified an insect problem to automatically alert the agronomist with a list of products that are in stock (via their ERP connection) plus calculate which would give the farmer the higher rebate, schedule a spray application (eg: in Slingshot), and log the observations in a specific FMS (eg: Fieldalytics) you could have multiple APIs stacked on, or you could use the MCP to connect to. It helps provide enhanced context plus enables the LLM to take actions— delivering an agent experience.

This is where LLMs seemingly unlock more value beyond basic question and answer. Effectively, MCP can make LLMs more capable.

Today, the number of problems that LLMs solve in ag are few. But MCP could help get LLMs to a point where they do unlock more value.

Earlier this year I wrote, 6 Ways I Would Use GenAI If I Were Still an Agronomist in Ag Retail, looking at specific ways singular individuals can use GenAI systems and tools. It focused on the basics. But if we see MCP prove valuable, and LLMs move from From Chatbots to Digital Teammates in an agentic way, there is going to need to be thoughtful consideration by ag leadership as to where to focus: AI Requires Bottom up Reasoning, Not Top Down.

AI Requires Bottom up Reasoning, Not Top Down

That means a focus on three things to start:

1. Define the Capability Pathway and the Problems to Solve
Treat LLMs like infrastructure, not features. What is the specific problem you’re solving—compliance reporting, internal knowledge retrieval, something within your supply chain? Then work backward: what is required to solve that? What roles does it impact? What component of the problem will it solve?

2. Build a Structured, Proprietary, and Connected Data Layer
LLMs are only as good as the data they access. What structured data do you already own or have? Where is it stored? How frequently is it updated? Connect internal silos (R&D, customer service, agronomy, logistics). Companies that find success with LLMs will treat their data as an asset— not just an input.

3. Identify Workflow Control Points
Where in your process does a decision get made? What tools do agronomists, salespeople, and ops leads use every day? That’s where AI should live. Don’t bolt on another platform—embed intelligence into existing systems at the point of action. Start with augmenting workflows, not replacing people. This can be as simple as asking your preferred platform provider about their strategy and influencing the functionality as it gets developed.

MCP could be a catalyst for moving LLMs ahead, and has the potential to alleviate some of the shortcomings that have been holding LLMs back.