What Is MCP and Is It Worth It? Part 1: Framing the Value
- Dave Findlay
- Oct 9
- 4 min read
Over the past few months, MCP has found its way into nearly every conversation I’ve had with CIOs and data leaders.
Vendors are promoting it. Engineers are experimenting with it. And CIOs are asking:
“Is this just another GenAI thing or is there real business value here?”
This two-part series is designed to answer that question. We’ll break down what MCP is (and isn’t), where it fits into the modern enterprise stack, and what value it can deliver, not just for developers, but for the business as a whole.
MCP Plainly Explained
MCP stands for Model Context Protocol. At a high level, it’s a structured way to define what tools (in tool-calling) an LLM can access, what inputs those tools expect, the context that will be helpful, and how to route user requests accordingly.
You can think of it as a standardized interface between natural language systems and business systems, much like REST APIs are an interface between multiple business systems.
If tool-calling is how LLMs get things done, MCP is how they understand what can be done and how to do it safely and repeatably. Think of it somewhat like an application server or business rules engine that handles the inputs and outputs necessary to make your agentic solution come to life.
Why It’s Gaining Attention
As teams move from GenAI pilots to production use cases, one thing becomes clear: free-form prompting, managing tool budgets in code, and injecting context manually doesn’t scale. Business users want consistency. Governance teams want control. Architects want composability.
MCP offers a framework for all three.
By defining a catalog of tools and actions, along with parameters, types, validations, descriptions, and examples MCP lets you:
Control what LLMs can do (and can’t do)
Encourage reuse and modularity across use cases
Improve accuracy by grounding prompts in formal context
Set the stage for agentic workflows
Why MCP Matters from a Business Perspective
Too many GenAI projects stall out after the pilot phase. Why?
Because early experiments are often built with brittle code, limited reuse, and little regard for scalability or governance.
MCP changes that.
By formalizing how LLMs interact with enterprise systems, MCP:
Reduces risk by limiting what LLMs are allowed to do
Lowers cost by making actions modular and reusable
Improves reliability by grounding prompts in structured context
Accelerates delivery by separating design from implementation
In other words, MCP isn’t just a developer convenience. It’s a strategic enabler for CIOs trying to scale GenAI safely, cost-effectively, and in a way that aligns with enterprise standards.
Without MCP, organizations face a different reality:
Increased risk from unmanaged tool use and ambiguous prompts
Slower development cycles due to one-off implementations and fragile pipelines
Higher costs from redundant or overly complex prompt engineering
Limited scalability as new use cases require bespoke integrations and governance patches
Put simply, skipping MCP might work for isolated pilots but it makes enterprise-scale deployment expensive, inconsistent, and hard to govern.
What It Looks Like in Practice
Let’s say you want to let users ask questions like:
“What was our average occupancy rate last quarter for my portfolio?”
Rather than leaving the LLM to figure everything out from a hardcoded list of tools in a Python script, MCP gives it:
A list of tools it can use (e.g., `runSQLQuery`)
A structured definition of what inputs that tool expects (e.g., natural language string, a validated SQL string, JSON formatted results, etc.)
Examples of how similar requests have been handled in the past
User context to resolve ambiguity like “my portfolio -> Ontario”
This is where design matters. A good MCP implementation acts like a bridge:
On one side: natural language questions
On the other: business logic, systems, APIs, and data assets
MCP connects the two, with guardrails.
Where It’s Being Used
MCP is a core concept in many modern cloud and data platforms. For example, Snowflake provides a managed MCP Server as part of its AI/ML product offering. It’s also appearing in custom enterprise architectures built on platforms like LangChain.
But it’s not a product. It’s a pattern.
And like all patterns, its success depends on:
The quality of your tool definitions
The consistency of your structure
The governance model you apply
Is It Right for Your Company?
That depends. If you’re experimenting with GenAI in isolated use cases, MCP will be overkill. But if you’re:
Trying to build scalable GenAI products
Supporting multiple user types and workflows
Embedding LLMs into production systems
Then MCP may offer the structure you need.
The key is to approach it not just as a technical artifact, but as a design and governance layer.
That’s where we’ll go in Part 2.
We’ll explore:
What makes a good tool definition
How to balance reusability and specificity
The ROI of investing in MCP-style design
And we’ll give you a checklist to help decide whether now’s the time to make that investment.
TL;DR
If GenAI is part of your enterprise roadmap, MCP might be the missing layer that makes it safe, reusable, and scalable.
It’s not just about what LLMs can do.
It’s about making sure they do the right things, the right way, every time.
Follow along for Part 2: Making the Business Case for MCP.
At Fuse, we believe a great data strategy only matters if it leads to action.
If you’re ready to move from planning to execution — and build solutions your team will actually use — let’s talk.
