At some point, you’ve likely welcomed a recent college graduate into your business. They’re smart, well-educated, and full of potential—but on day one, they have little understanding of your company’s unique processes, culture, or goals.
Large language models (LLMs) are much the same. They carry vast general knowledge yet lack the specific context that makes them immediately valuable to your organization. Just like new hires go through the onboarding ropes, LLMs need structured access to your business’s data, tools, and workflows to become truly useful.
That’s where Model Context Protocol (MCP) comes in. MCP enables communication between AI applications, AI agents, applications and data sources. The protocol has quickly moved from an emerging standard to a strategic enabler, and the conversation we’re having with our clients has shifted from technical architecture to practical application.
MCP is not just another integration layer. It’s a way to unlock latent value across your organization by connecting AI agents with the systems, data, and workflows that drive outcomes. The real opportunity lies in how you apply MCP.
Start with what and why
Let’s be honest, there’s no shortage of MCP primers out there. Most of them walk you through the architecture: hosts, clients, servers. That’s fine, but it’s not where the real value is. The real question isn’t, “How does MCP work?” It’s “What can I do with it?” and “Why does it matter to my business?”
When we talk about MCP, I try to steer the conversation away from the architecture and toward the outcomes. What problem are you solving? Why is MCP the right tool to achieve your goals?
A Midwest health system we worked with wanted to personalize treatment for patients with hypertension, using the vast troves of data stored in their electronic health records (EHR). The strategic hurdle wasn’t just accessing the data, it was giving access securely, at scale, and in a way that respected compliance boundaries across thousands of patient encounters.
With MCP, we were able to connect AI agents to a rich EHR data model that included vitals, medications, comorbidities, lab results, and even nuanced metrics like ejection fraction readings. MCP serves as the structured conduit, enabling the AI to interact with nearly 800 patient features per encounter without compromising privacy or requiring custom integrations.
The predictive accuracy has enabled clinicians to tailor treatment plans with greater precision, according to our client. Patients have gained an estimated 100 additional days of life, and the system saw $2,000 in annual healthcare savings for 20% of its hypertension population.
Similarly, a national convenience store chain used MCP to connect AI systems with real-time data on customer movement, promotional engagement, and inventory shrinkage. No retraining models. No custom APIs. Just a scalable model for improving store performance.
MCP isn’t just a bridge between systems. More vitally, it connects strategic intent with measurable outcomes.
Guardrails for autonomy and accountability
As we move toward agentic AI—models acting like digital employees—autonomy without structure is risky. You wouldn’t let a new hire run wild with sensitive data or make decisions without oversight, and the same goes for AI.
One major challenge is idempotency, or the ability to perform the same operation repeatedly with consistent results. Most LLMs aren’t idempotent. Ask one to write an email five times, and you’ll get five different versions. That’s fine for creativity, but not for processing payments or executing compliance workflows.
MCP introduces guardrails to standardize how agents interact with internal systems, ensuring repeatable, reliable outcomes even if the AI’s internal reasoning varies. That’s critical in regulated industries like healthcare, finance, and government.
We saw this play out with a Middle East government statistics agency. They had decades of data buried in legacy systems. MCP enabled a voice-powered AI interface that could query massive datasets in Arabic and English. What used to take weeks now takes seconds, and more importantly, decision-makers now have timely, contextual intelligence at their fingertips.
Strategic implementation: build once, scale everywhere
Here’s the thing: MCP isn’t about building one-off solutions. It’s about creating frameworks that can be reused across departments and use cases.
To apply MCP effectively, organizations must think in the following terms:
- Reliability and repeatability: MCP enforces structured communication, making AI agents more predictable and trustworthy.
- Scalability and ecosystem growth: With a unified API layer, MCP simplifies deployment and integration, accelerating innovation.
- Safety and control: MCP ensures AI agents operate within defined boundaries, protecting sensitive data and maintaining enterprise integrity.
We worked with a global healthcare technology provider that wanted to simplify complex medical terminology for patients. Instead of building a narrow solution, we used MCP to create a reusable framework that could be extended across departments.
AI agents can securely access structured medical data and terminology libraries, apply consistent translation logic, and tailor outputs for patients, clinicians, and administrative staff.
That same protocol-driven infrastructure was later adapted for internal training, multilingual documentation, and voice-assisted navigation of clinical systems. MCP made it possible to replicate success without reinventing the wheel.
That’s what strategic implementation looks like—turning isolated wins into enterprise-wide transformation.
The road ahead
MCP is more than protocol, it’s a strategic enabler. It gives AI agents the structure they need to interact with enterprise data and tools. This means businesses can unlock new efficiencies, reduce development cycles, and build a thriving ecosystem of interoperable AI solutions.
The full potential is still unfolding, but for companies serious about AI, working with partners that understand how to apply MCP can be foundational. With the right guardrails in place, AI can be creative and compliant, autonomous, and accountable.
Just what you’d expect from any employee helping move your business forward.
Juan Orlandini is CTO North America of Insight Enterprises.