Before diving in, it is useful to introduce a simple framing. When organisations adopt MCP or agentic systems, they tend to encounter three distinct perspectives for creating value. These are not maturity stages or strategic choices. They are conceptual lenses – different ways of understanding what MCP and Agentic AI unlocks:

1. The Interface Perspective (Dimension 1): how MCP enables a new generation of intelligent user interfaces for knowledge work.
2. The Integration Perspective (Dimension 2): how MCP allows agents to orchestrate capabilities across siloed systems and bypass traditional integration bottlenecks.
3. The Composability Perspective (Dimension 3): how MCP provides modular building blocks that enable standardised, reusable agent architectures.

Each perspective stands independently. An organisation may explore any one of them without touching the others. With this framing in place, we can now examine each perspective – and the business value it represents – in greater detail.

The most immediate value of MCP is the transformation of the end user interface.

Consider the case of Block (Square). They developed an internal platform (called “Goose”) using MCP to connect fragmented tools – Jira, GitHub, Snowflake, and HR systems – into a single conversational interface. They report that engineering teams save 50–75% of their time on common tasks such as triage and code migration.

This efficiency does not come from AI inventing new processes. It comes from removing the friction of context-switching between dashboard silos.

The shift of the gatekeeper
To understand the staying power of this, consider the history of digital interfaces:

• In the 90s, the Browser and Search Engines were gatekeepers to value.
• In the 2010s, the App Stores were the gatekeepers.
• Today, the Chat Window is becoming the new entry point.

MCP is the plugin standard that allows that Chat Window to execute actions beyond simple text generation. Pure Chat Windows are very low bandwidth UIs, but this is also changing. New developments like MCP UI (where the chat returns interactive widgets) suggest a return to applet-like functionality, but with higher intelligence.

Because chat agents are here to stay, this dimension justifies the investments – you can build products or even entire companies around it.

Importantly, Dimension 1 stands fully on its own. Organisations can deploy personal productivity agents without creating any system-to-system integrations or adopting a composable architecture. Governance is also manageable because these agents can rely on “identity mirroring” – inheriting the user’s own IAM permissions. This makes Dimension 1 the lowest-risk and fastest-adopted perspective on MCP.

The second dimension addresses a legacy problem that has plagued enterprises for decades: Integration Debt.

Siloed data and rigid API contracts often paralyse cross-functional automation. Fifteen years ago, the industry attempted to solve this with Robotic Process Automation (RPA). The premise was to bypass IT bottlenecks by automating the User Interface.

RPA worked for a time, but it was structurally brittle. It relied on screen coordinates and DOM elements. When the UI changed, the automation broke, leading to escalating maintenance costs that eroded ROI.

MCP functions as a structural evolution of RPA.

It shares the same aggressive integration strategy – bypassing the need for complex middleware projects – but replaces the method. Instead of automating Pixels (which are fragile), MCP automates Capabilities (APIs and UI functions).

It’s important to clarify that MCP is not “agentic” by default. MCP is simply an interface layer – it exposes capabilities. Whether you attach a deterministic script, a conversational assistant, or a fully autonomous agent on top of those capabilities is a separate architectural choice. In this article, I focus specifically on agentic MCP systems because they highlight the governance issues most organisations will face.

Again, this perspective exists independently. Some organizations will adopt Dimension 2 to address a single high-value operational bottleneck – without ever deploying personal assistants or investing in composable agent architectures.

We are seeing a trend where IT departments use MCP – much like internal API gateways – to streamline the politics of integration. By requiring departments to expose standard MCP servers, organisations bypass the need for bespoke, point-to-point integration projects.

The third dimension is often misunderstood as a “Marketplace for Digital Coworkers.” In practice, the real value is technical composability, and that is not unique to agentic use cases.

In a fragmented organisation, multiple teams often attempt to build AI agents for their specific verticals simultaneously. If they all build from scratch, the result is fragmentation, fatigue, and inconsistent security.

MCP promotes Composable Architectures. Organizations can organize components in layers:

• Interface Layer: how users interact with the agent
• Cognitive Layer: how the agent thinks and acts
• Capability Layer: what the agent is able to do

Instead of building a monolithic “Supply Chain Agent,” teams compose it from these certified parts. This solves the “Empty Factory” problem. It turns the creation of agents from a bespoke art project into a standardised assembly line, leveraging the same governance patterns IT already uses for APIs.

Important: Dimension 3 is not about enabling infinite agent configurations. In reality, most organisations only need 5–10 recurring agent compositions. The purpose of a composable architecture is not to build more agents – it is to build them in a way that is maintainable, governable, and secure once the number of agents grows. It is the strategic choice you make when you believe that the Total Cost of Ownership (TCO) of maintaining dozens of bespoke agents will exceed the upfront investment in a standardised agent factory.

A useful way to understand Dimension 3 is through its internal evolution. Once an organisation chooses to pursue composability, it naturally progresses through three stages:

• Code Blueprints – teams clone source code templates; drift is high; governance is weak.
• Boxed Agents – central teams maintain standard “ready to deploy” compositions; updates and governance improve; performance is tuned per instance, but the underlying code remains the same.
• Modular Composable Architecture – teams assemble agents from standardised MCP modules; governance applies to modules, not agents; scale becomes sustainable.

The most significant barrier to MCP adoption is not technical; it is the gap between enterprise governance models and the nature of Agentic AI.

Enterprises are designed to procure and audit Tools. A tool is deterministic: Input X yields Output Y. It can be certified and subjected to standard operational risk assessments.

Agentic AI functions less like a tool and more like a dynamic Digital Employee. The tool and the process are inseparable – just like with RPA. The challenge is that this “Employee” is probabilistic. It may produce different outputs given similar inputs based on context or configuration.

This creates inevitable friction with existing compliance frameworks:

• The Audit Challenge: Standard Data Protection Impact Assessments (DPIAs) or Operational Impact Assessments (OIAs) struggle with software that decides at runtime which data sources to inspect or how it acts on it.
• Access Control: Security teams must move beyond simple API access to granular permissioning. We must classify MCP tools as Read-Only versus Destructive. An agent may be permitted to read Jira tickets, but should it be autonomously permitted to DELETE records in a production database?
• The “Human-in-the-Loop” Lie: Relying solely on human oversight for high-volume workflows creates a false sense of security. If an agent operates with 99% accuracy, human operators tend to disengage, allowing the 1% error rate to pass undetected.

Each perspective introduces a different governance burden.

• Dimension 1 relies on inherited user identity.

The business case for MCP depends on whether you establish the required governance processes, define a suitable security model, and align your organisation with the dynamic nature of agentic systems.

These three dimensions are not a roadmap or a maturity curve. They are independent perspectives — different ways MCP can create value. An organization may benefit from one without touching the others.

• For Dimension 1 (Intelligent Assistance), the technology is ready. Connecting internal data to chat interfaces yields measurable productivity gains with manageable risk.
• For Dimension 2 (Integration Bypass), the value is high, but so is the governance burden. Most organisations are not yet structurally prepared for probabilistic system integration.
• For Dimension 3 (Composability), the investment is long-term. It is about preventing architectural drift and maintaining control as the number of agents grows. The key question is not if you will need this, but when. The tipping point arrives when the cost of maintaining your existing bespoke agents exceeds the investment in a common platform.

MCP is not a magic fix for enterprise architecture. It is, however, a pragmatic strategy to unlock value from existing imperfect systems. It does not clean up the legacy landscape; it simply makes that landscape usable and economically productive for the next generation of automation.