Why Should IT Leaders Be Thinking About Model Context Protocol?

Last November, Anthropic rolled out the Model Context Protocol (MCP), which initially attracted muted interest. The company tucked the news into a blog post, calling MCP an open standard meant to “help frontier models produce better, more relevant responses.”

But as developers learned more about MCP, it became clear how powerful it was. In a few months, companies like OpenAI, Google, and Microsoft adopted the standard. This fueled interest in MCP, as the growth resembled a red-hot consumer app, not a developer infrastructure tool.

The GitHub repository for MCP has quickly turned into a vibrant community. Currently, there are more than 64,500 stars and about 7,500 forks. Then there are the thousands of servers that have popped up on various websites.

Such momentum is rare for developer infrastructure. Yet it shows the importance of MCP, as it has become referred to as the “USB C for AI apps.”

So let’s see why this open standard has become so popular and how IT leaders should think about it.

The Benefits of MCP

Before the introduction of MCPs, building advanced generative AI or agentic systems was a painstaking process. Each large language model (LLM) required custom integration with every tool or data source it used. This created what’s known as the “MxN problem.” This where M models have to be manually connected to N different tools.

For example, if you use three different LLMs to work with ten applications, you will need to build 30 separate integrations. Not only will this require significant engineering resources, but the codebase will be difficult to maintain as the tools, APIs, and models evolve.

But with the MCP standard, the process is significantly improved. It provides for two important capabilities: context and tool use with LLMs. This allows for not only more relevant responses but improved accuracy and productivity.

For example, with context an AI application can access a wide range of publicly available data sources, say for weather or financial data. MCPs also access private data sources like Slacks or Jira tickets.

In terms of tool use, an MCP can carry out actions like CRUD tasks for databases, scheduling events or reminders, or updates for CRMs or ERPs.

Besides providing standardization for context and tool use, there are other advantages with MCP. One is security, as it supports OAuth-based authorization. Next, models are not tightly coupled with tools or data sources.  In other words, when APIs change or a new tool is adopted, there is no need for major rewrites.

MCP also helps to improve governance and compliance because of the centralization of tool use and data flows. This makes it easier to enforce policies and audits.

In light of these advantages, it should be no surprise that MCP has turned into a highly popular system for building generative AI and agentic applications.

Challenges of MCP

MCP still needs much work to make it more stable and mature. The UIs are often clunky and unintuitive. To improve security, MCPs should also have strongly typed approaches to minimize potential attack vectors. Just as important is fine-grained authorization. For example, it should be possible to authorize an MCP server or agent only for specific actions.

Discovering MCPs remains a problem as well. What’s needed are registries to validate and certify servers, similar to how app stores work.  These registries can serve different verticals, like IT, security, and finance. Enterprises are likely to develop internal registries to provide even more control.

Finally, MCPs may have broader implications, even threatening business models. For example, these systems could lower daily active users (DAUs) for web applications and mobile apps. The reason is that AI agents will leverage MCPs to carry out actions, which means less need for human users to visit the platforms.

Security as a Foundation

MCPs allow for much faster innovation. This is especially important as enterprises face mounting pressure to show tangible results from their AI investments. However, the drive for speed must not come at the expense of security and compliance. Cutting corners in these areas can create significant risks, given that MCPs not only access sensitive data but can also take direct actions with it.

A MCP implementation should embed governance, logging, and auditing into every layer. Policies need to clearly define who can authorize agents, what actions they are permitted to perform, and how those activities are monitored. Granular authorization, combined with continuous oversight, reduces the potential for misuse while ensuring the transparency required for compliance.

Conclusion

MCP is rapidly becoming a cornerstone for building the next generation of generative AI and agentic systems. For IT leaders, MCP represents both an opportunity and a responsibility. There is the opportunity to unlock new efficiencies and capabilities, and the responsibility to implement it with the right guardrails in place.

In the long run, enterprises that treat security and compliance as integral, not optional, will be best positioned to capture MCP’s full value. By balancing innovation with strong governance, IT leaders can ensure that their AI initiatives are not only powerful and transformative, but also trustworthy, sustainable, and resilient.