Over the last year, I have been watching the AI conversation shift.
For a while, most of the attention was on the model itself: which model was smartest, fastest, cheapest, or best at writing code, summarizing documents, or drafting content. That still matters. But in my view, the next meaningful stage of AI adoption will be less about the model in isolation and more about what the model can securely connect to.
That is where MCP servers come in.
MCP stands for Model Context Protocol. It is an open standard that provides AI systems with a common way to connect to tools, data sources, business systems, and workflows. A simple way to think about it is this: if a chatbot or AI assistant is the user interface, an MCP server is one of the controlled connection points that lets that assistant safely access something useful.
That “something useful” might be a document repository, a customer relationship management system, a payment platform, a software codebase, a ticketing system, a calendar, a cloud service, or an internal database.
For businesses, this is an important distinction. A general-purpose AI tool can answer questions based on what it already knows or what a user pastes into the conversation. An AI tool connected through MCP can potentially work with the systems a business already uses. That moves AI from being a helpful standalone assistant to becoming part of the organization’s operating layer.
What MCP Actually Does
MCP was introduced by Anthropic in 2024 as an open standard for connecting AI assistants to the systems where business data and tools live. Since then, it has been adopted by a growing number of AI platforms, developer tools, and enterprise technology providers.
In practical terms, MCP creates a common structure for three things:
First, it can expose resources. These are pieces of information an AI assistant can read, such as files, API responses, documentation, database records, or other structured content.
Second, it can expose tools. These are actions the AI assistant can ask to perform, such as creating a support ticket, searching a repository, updating an issue, checking a payment status, or triggering a workflow.
Third, it can expose prompts. These are reusable instructions or templates that help guide the AI system through a particular task.
The important point is that MCP does not magically make AI “understand the business.” Instead, it creates a standardized way for AI applications to request context and actions from approved systems. That standardization is what makes it interesting.
Without standards, every AI integration becomes a custom project. One vendor has its own way of connecting to tools. Another vendor has a different approach. Internal teams end up building one-off integrations that are hard to maintain and harder to govern. MCP is an attempt to reduce that fragmentation.
How MCP Servers Are Being Used Today
Right now, MCP is most visible in software development and technical operations. That makes sense. Developers are early adopters of AI tools, and many of the first valuable use cases involve connecting an AI coding assistant to repositories, documentation, issue trackers, and cloud infrastructure.
GitHub, for example, has an official MCP server that allows AI tools to connect directly to GitHub. This can help an assistant read repositories, search code, work with issues and pull requests, and support development workflows through natural language interaction.
Stripe also provides MCP support so AI-enabled development tools can connect to Stripe. That means developers can work with payment-related documentation, APIs, and account-specific capabilities in a more integrated way.
Cloudflare provides MCP servers that allow AI tools to read configurations, manage services, and automate actions across Cloudflare accounts. Atlassian has also introduced remote MCP server capabilities to connect AI assistants to Atlassian tools, with authentication and permission controls.
These are not abstract examples. They show where the market is going. Major software platforms are preparing for a world where AI agents are not just answering questions; they are operating across systems.
For many organizations, the first wave of MCP adoption will probably appear in areas like:
Software development and DevOps
IT support and managed services
Knowledge management
Customer support operations
Sales and account management workflows
Finance and billing operations
Compliance, reporting, and documentation
This does not mean every company should immediately connect AI agents to every business system. In fact, that would be a mistake. The better starting point is to identify controlled use cases where the value is clear and the risk can be managed.
For example, an internal AI assistant that can read approved documentation and summarize support procedures is a lower-risk starting point. An AI assistant that can make infrastructure changes, move money, or update customer records requires a much higher level of governance.
Why MCP Matters for Digital Transformation
At Swinmark, we tend to look at technology through a strategy-first lens. The question is not “What can this tool do?” The better question is “What business capability does this make possible?”
MCP matters because it could become a foundational layer for AI-enabled digital transformation.
Many organizations have already invested in cloud systems, CRMs, ERPs, support platforms, analytics tools, and custom applications. The challenge is that these systems often remain fragmented. People still spend too much time searching, copying, pasting, reconciling information, and translating business intent into system-specific tasks.
MCP has the potential to reduce that friction.
Imagine asking an internal AI assistant: “Which customers have open implementation issues, unpaid invoices, and no scheduled follow-up meeting?” Without system access, the assistant can only provide general advice. With well-governed access via MCP servers, it could query the CRM, ticketing system, billing platform, and calendar and produce a useful operational summary.
Or imagine a managed services scenario where a client asks, “Why was our website slow yesterday afternoon?” An AI assistant connected to approved monitoring tools, deployment history, incident logs, and CDN configuration data could help the support team identify likely causes faster.
The real value is not simply automation. There is better coordination between information, systems, and decision-making.
The Future: From Chatbots to Business Agents
In the near future, I expect MCP servers to play a major role in the shift from chatbots to business agents.
A chatbot responds. An agent acts.
That distinction is important. When AI systems can connect to tools and data sources, they can begin to complete multi-step work. They can check information, compare options, draft a recommendation, request approval, and then trigger an action.
This could reshape many common business workflows.
In project management, an AI agent could review project notes, identify blockers, update tasks, and prepare a status report.
In sales operations, it could summarize account activity, identify stalled opportunities, draft follow-up emails, and recommend next actions.
In customer support, it could review a ticket, search the knowledge base, check account history, suggest a resolution, and escalate when confidence is low.
In IT operations, it could review alerts, compare recent deployments, check infrastructure status, and recommend remediation steps.
In compliance-heavy environments, it could gather evidence, prepare audit trails, and flag missing documentation.
The future impact will be significant, but I do not believe it will be evenly distributed. Organizations with clean systems, good data governance, clear processes, and strong security practices will benefit faster. Organizations with fragmented systems and unclear ownership will struggle, as MCP does not eliminate the need for good architecture. It exposes the quality of the existing architecture.
The Security Question Cannot Be an Afterthought
There is another side to MCP that business leaders need to understand: it increases the importance of security, governance, and system design.
When an AI assistant can access business tools, it also creates new risk. What data can it read? What actions can it perform? Who approved those permissions? How are actions logged? Can a malicious prompt influence the assistant into doing something it should not do? Can an untrusted MCP server pose a supply chain risk?
Security researchers have already identified risks in MCP implementations, including tool poisoning, prompt injection, authentication weaknesses, and unsafe access patterns. These risks do not mean MCP should be avoided. They mean it should be implemented with discipline.
For business leaders, I would frame MCP governance around a few practical principles.
Start with read-only use cases before allowing write actions.
Use least-privilege permissions so that each MCP server can access only what it truly needs.
Require human approval for high-impact actions, especially in finance, infrastructure, legal, HR, and customer data workflows.
Keep audit logs of what the AI accessed, what it recommended, and what actions were taken.
Use trusted MCP servers from reputable vendors where possible.
Treat third-party MCP servers like any other software supply-chain dependency.
Test for prompt injection and unsafe tool behaviour before production deployment.
In other words, MCP should not be treated as a shortcut around IT governance. It should be treated as a new integration pattern that requires the same level of architectural thinking as APIs, cloud services, identity management, and data platforms.
What Business Leaders Should Do Now
For most small and mid-sized organizations, the right move is not to rush into a large MCP implementation. The right move is to start learning, mapping, and prioritizing.
I would begin with three questions.
Where do our teams lose time moving information between systems?
Which systems contain the context our people need to make better decisions?
Which workflows would benefit from AI assistance, but only if the AI had secure access to approved business data?
Those questions help move the conversation away from hype and toward business value.
A good first project might be an internal knowledge assistant connected to approved documentation. Another might be a development assistant connected to a code repository and issue tracker. Another might be a managed services assistant that helps support teams interpret monitoring alerts and client history.
The best use cases will share three characteristics: clear value, limited risk, and measurable outcomes.
My View
I do not see MCP as just another AI acronym. I see it as part of a larger shift in how businesses will connect intelligence to operations.
The first generation of business AI was largely about content generation and summarization. The next generation will be about context, workflow, and action. MCP servers are one of the standards making that possible.
But the winners will not be the companies that connect AI to everything as quickly as possible. The winners will be the companies that connect AI to the right things, with the right controls, for the right business reasons.
That is where strategy matters.
MCP is a technical protocol, but its impact will be organizational. It will force companies to think about their systems, permissions, data quality, operating processes, and risk tolerance. It will reward businesses that have invested in clean architecture and punish those that have allowed fragmented systems to grow unchecked.
For leaders exploring AI, MCP is worth understanding now. Not because every organization needs to deploy MCP servers immediately, but because the direction is clear: AI is moving closer to the systems where work actually happens.
And when AI becomes connected to the work, the real transformation begins.
References
- Anthropic. “Introducing the Model Context Protocol.” November 25, 2024.
- Model Context Protocol documentation. “Understanding MCP Clients” and related MCP architecture documentation.
- OpenAI Agents SDK documentation. “Model Context Protocol.”
- Microsoft Copilot Studio documentation. “Extend your agent with Model Context Protocol.”
- GitHub. “GitHub MCP Server.” Official GitHub repository.
- Stripe documentation. “Model Context Protocol (MCP).”
- Cloudflare documentation. “Cloudflare’s own MCP servers.”
- Atlassian. “Extend Atlassian into any AI assistant using MCP.”
- Hasan, Mohammed Mehedi, Hao Li, Emad Fallahzadeh, Bram Adams, and Ahmed E. Hassan. “Model Context Protocol (MCP) at First Glance: Studying the Security and Maintainability of MCP Servers.” arXiv, 2025.
- Radosevich, Brandon, and John Halloran. “MCP Safety Audit: LLMs with the Model Context Protocol Allow Major Security Exploits.” arXiv, 2025.
- Zhou, Huijun, Xiaohan Zhang, Haozhe Zhang, Haoyang Zhang, Mi Zhang, and Min Yang. “A First Measurement Study on Authentication Security in Real-World Remote MCP Servers.” arXiv, 2026.
