It’s early days, to be sure, for Anthropic’s Model Context Protocol (MCP). But it is definitely time for CIOs to put this smoking hot, open-source protocol for connecting AI applications to databases, web services, application systems, and local resources on their radar.
Today, MCP’s rabid fan base is confined to the AI development community, where its presence is expanding quickly. Indeed, most other developers haven’t heard of it yet. But CIOs should expect that to change quickly as MCP aims to be a universal connector.
What is MCP?
In a nutshell, MCP uses a client-host-server architecture: The AI application side acts as a “client” or “host,” and external systems and apps expose MCP “servers” that offer resources, tools, or data. The AI tooling world touts MCP as a promising standard for “AI-native integration.” That doesn’t mean APIs are down for the count — they’re still No. 1 in the integration game. Watch MCP in action, and you’ll see it wraps existing APIs. In short, MCP is more of a standardization layer on top of APIs for AI contexts, not a replacement for integration at large.
So why not just stick to using APIs, you might ask? There are many reasons, but a primary driver is the next-gen enterprise AI system: agentic AI. Traditional APIs still matter in agentic systems, but they require more dynamic, context-aware, and persistent connections than static APIs alone. MCP is a key enabler of agentic AI.
If you haven’t heard of agentic AI or AI agents, it refers to autonomous AI systems that pursue goals over time by reasoning, remembering, and acting to achieve desired outcomes. These AI assistants and agents are designed to independently complete business and personal processes, somewhat like the AI systems driving autonomous vehicles now, even though they are very different technologies.
This drive for autonomy and efficiency is mirrored in the way businesses are adopting MCP to integrate internal systems and data. Kevin McDonald, senior consultant at Hylaine, a tech consulting firm, highlighted one prominent example: the development of custom MCP servers designed to connect company knowledge bases and CRMs. These servers enable employees to get quick answers about policies and customer data, improving operational efficiency.
“So far, it [the MCP server] has performed better than expected: It started out as a development tool, and has grown into a technology that forms the backbone of agentic systems,” McDonald said.
Case Study: MCP in Hospitality
Several midsize independent hotels have become early adopters of Apaleo’s MCP server, a platform designed to simplify the integration of AI agents with hospitality systems by eliminating the need for custom coding. Apaleo, a hospitality technology company, said the MCP server is used by more than 2,000 properties worldwide, including hotels owned by CitizenM (part of the Marriott group), easyHotel, Numa Hotels, and Limehome.
Stephan Wiesener, CTO and Co-Founder of Apaleo, said early adopters are using AI agents connected to the Apaleo MCP server to drive efficiencies across operational workflows, such as revenue management, corporate sales, and finance .
-
In revenue management, AI agents autonomously access historical booking data through MCP to analyze guest behavior, spot booking trends, and recommend pricing adjustments.
-
In corporate sales, agents produce live reports that identify top-performing business accounts and monitor performance, enabling hotels to shift from static discounts to volume-based contracts.
Wiesener said some hotels have started using LLM models and the MCP server to generate automated briefings each morning, detailing revenue, arrivals, departures, and occupancy across properties. These summaries are delivered to on-site teams without any manual input, saving staff work time and ensuring more consistent operational oversight, according to Wiesener. Hoteliers are also trialing agent systems via MCP in the end-to-end automation of accounts receivable. The aim is to use AI agents to retrieve balances, classify accounts, send follow-ups, and even recommend write-offs. Before the advent of agentic AI and MCP servers, such workflows often required heavy custom coding and cross-department coordination.
The autonomous hotel
In addition to automating specific tasks from start to finish, some of the hotels are beginning to assign AI agents broader organizational roles. The roles involve responsibilities that expand the agents’ ability to take contextual actions — tasks based on specific circumstances or changing information — across multiple systems using MCP, Wiesener said.
Most notably, these hotels can now experiment, adapt, and scale AI internally without relying on developers. This is a major leap toward the industry’s next era: the autonomous hotel.
The ‘USB-C of LLMs’?
The question then becomes whether early successes such as these can be replicated and amplified across industries.
Joseph Ours, partner and AI solutions director at Centric Consulting, described the current industry impact of MCP as occupying a “consequential middle ground.”
“It’s neither the transformative revolution some predicted nor the mere incremental tooling others dismissed,” Ours said, referencing an online meme that there are more MCP developers and servers than there are customers.
Implementation matters
In truth, MCP is following the ups and downs typical of new technologies, including the hordes that try to cash in early, said Ours, an early contributor to FastMCP, which is now the de facto standard for Python-based MCP servers. The protocol itself is a good standardization approach, but its performance and reliability can vary significantly.
“While the protocol is solid, many MCP servers are ‘vibe coded‘ and have varying degrees of quality built into the actual MCP server,” Ours said.
“Two different MCP servers performing the same function may complete in 15 seconds or over 4 hours, depending on the architecture used for the actual calls behind the protocol,” he said, adding that the key insight for IT and business leaders is that implementation matters.
Still, while caution is certainly smart at this stage, there’s no denying that MCP is progressing quickly.
Xiangpeng Wan, product lead at NetMind.AI., described MCP as the “USB-C of LLMs ,” recounting its rapid adoption by the major AI technology providers.
“In March 2025 OpenAI announced it would integrate MCP into the ChatGPT desktop app and its Agents SDK,” he said. “Then in April, it was Google DeepMind saying its Gemini models would support MCP as well. Microsoft and others not only back the protocol but have also released servers like Playwright-MCP so AI assistants can automate web actions through a browser.” All of which points to MCP’s emergence as the standard for connecting LLMS to external data.
Momentum is growing for what is essentially a disruptive approach to accessing software services, agreed Mohith Shrivastava, principal developer advocate at Salesforce. AI developers see it as a better and faster way to connect AI to the information and tools it needs to be semi- or fully autonomous in completing its tasks.
“For decades, we’ve accessed software services through websites and apps, clicking buttons and navigating menus to achieve a business goal. MCP disrupts this model by creating a universal bridge to these same services through natural language,” Shrivastava said.
In the end, it may be user expectation that drives MCP into mainstream demand status.
“Instead of logging into a specific application, a user can now accomplish the same task by simply having a conversation within their preferred AI agent, whether it’s ChatGPT, Claude, Slack, or a specialized enterprise agent,” Shrivastava explained. “This allows the underlying software service or tool to be accessed in a more intuitive, efficient, and integrated way.”
In other words, consumers and business users won’t have to learn to use any given tool; they’ll simply state the outcome that they want. Such ubiquitous ease of use will likely win over even the most reluctant consumers, indicating a bright future for MCP.
“So far, so good. Usually, when a protocol gets to this level of adoption that MCP has, it is tough to unseat,” said Tom Taulli, author of the AWS Certified AI Practitioner (AIF-C01) Study Guide and a consultant for AI deployments.
