Model Context Protocol (MCP): A Framework for Connecting AI and Business Systems

Imagine asking your AI assistant: “Is there a link between our customer traffic sources and revenue last week?”

You don’t need to check dashboards, export files, or reconcile reports. The assistant responds directly, using data from your analytics, CRM, and social platforms as if they were part of the same conversation.

This is the promise of Model Context Protocol, or MCP. This new open standard aims to make AI assistants useful not just as general knowledge tools, but as live interfaces to your business systems.

It sounds promising, but there’s a lot to consider. In this longer read, we look at where MCP stands today, the risks and opportunities it presents, and how this early-stage technology might evolve.

Why AI Needs Access

AI assistants today, whether chat-based tools like Claude and ChatGPT, embedded agents, or browser-based copilots, are excellent at reasoning and language, but are isolated from your data. They don’t have native access to the tools your team uses every day. That limits their value.

You might ask an AI to analyse performance for a financial quarter, but end up manually downloading reports, formatting CSVs, and hoping it can piece together a picture from partial inputs. It’s clear that this isn’t automation, but rather delegating tasks to a machine.

Meanwhile, most businesses now rely on a wide range of digital systems, including analytics platforms, customer relationship management (CRM) tools, content collaboration suites, financial software, and internal messaging systems. Each of these systems has its own data structure, access model and API conventions. Connecting them to AI assistants has typically required custom development or a patchwork of middleware solutions.

What MCP Offers

Model Context Protocol introduces a shared way for AI models to connect with business systems. The aim is to reduce friction between tools and intelligence by defining a standard interface for access and interaction.

Anthropic describes MCP as a kind of USB-C for AI. The analogy is helpful, but not entirely accurate. USB-C is a hardware standard; once a device supports it, it works with any compliant cable or charger. MCP, by contrast, is a software protocol. It requires an AI assistant, a server, and a business system to all speak the same language and to keep speaking as APIs, security models, and endpoints change.

A more fitting comparison might be early web APIs or RSS feeds: open, promising, and powerful when maintained, but prone to breakage and drift. MCP setups often involve a three-way dependency: the assistant, the server, and the connected tool. If any part breaks, for example, the server isn’t updated after a platform changes its API, the entire connection fails.

Today, configuring MCP typically requires technical expertise and developer effort. That may not always be the case. Tooling, templates, or vendor support could eventually make it more universal and accessible, but for now, that future remains speculative.

How it Works

MCP structures its connections around three core components:

• Tools define the operations that can be performed, such as retrieving data, submitting a query, or updating a record.
• Resources represent available data entities and schemas, providing the assistant with information about the types of data that exist and their corresponding structures.
• Prompts are instruction templates that guide how the assistant should interact with tools and resources. These often include a few worked examples, known as few-shot prompts, which show the assistant how to perform a task by example, rather than just relying on a description.

In theory, these building blocks allow an assistant to treat your internal systems not as opaque endpoints, but as structured, queryable interfaces. In practice, much depends on the quality and specificity of the prompts, which are often handcrafted and fragile. There is no universal standard for how these elements are defined, and implementations vary significantly across servers.

Progress and Practical Limits

The protocol is open-source, and several early server modules are available for common platforms, including spreadsheet tools, databases, and local file systems. Some early users have reported promising results, but these are anecdotal and have not yet been substantiated by formal case studies or business metrics.

Adoption has accelerated significantly since early 2025. Major tech companies have now committed to supporting the standard. In March 2025, OpenAI CEO Sam Altman officially announced MCP support across OpenAI’s products. Google DeepMind confirmed MCP support in upcoming Gemini models in April 2025, with CEO Demis Hassabis describing it as ‘rapidly becoming an open standard for the AI agentic era’. Microsoft has partnered with Anthropic to create an official C# SDK and released Playwright-MCP for browser automation.

Technical setup is still complex. Many services lack working MCP servers. Documentation is fragmented across projects, and installation typically requires developer time and effort. Even where working connections exist, underlying API rate limits and caching often mean that “real-time” data is delayed or stale.

The current ecosystem also depends heavily on individual contributors and small open-source projects. If a server breaks or is no longer maintained, the integration may silently fail, and enterprise requirements add further complexity. MCP does not yet provide formal support for permissioning, role-based access, audit logs, or compliance features, all of which are critical in regulated or security-conscious environments.

Where MCP Can Deliver Value

Despite these constraints, MCP is already enabling useful workflows in scenarios where data is structured and access needs are predictable.

In marketing, it can unify performance data from web, social and advertising platforms. In sales, it can combine CRM data with financial and communications logs to provide a more comprehensive view of pipeline health. Operations teams can use it to surface insights from project trackers and dashboards without manually exporting or reformatting data.

That said, these benefits are unevenly distributed. MCP is better suited for high-volume, well-structured data flows than for ad hoc, unstructured, or qualitative tasks. Teams relying on tools with weak APIs, proprietary data formats or limited documentation may face significant implementation hurdles and integrating legacy systems remains a hands-on task.

Even in ideal conditions, there’s a trade-off. It looks like many current integrations rely on community-maintained MCP servers, which lack a formal support model. If these break or are no longer maintained, AI access to critical data can fail, even when the underlying systems are working perfectly.

Strategic Considerations

It’s also worth noting that while MCP is an open standard, it is not a neutral one. Anthropic created and maintains it, which gives the company strategic control over its evolution and positioning, particularly as context-aware AI becomes more commercially valuable.

That is not inherently negative. But it does raise questions about long-term governance, support, and the role of other vendors. The competitive landscape has shifted dramatically. Microsoft, Google, and OpenAI have all adopted MCP, suggesting it may become a de facto standard. Amazon remains the notable holdout among major cloud providers, continuing to develop its proprietary agent infrastructure.

For MCP to take off, it probably needs one of three things: stronger support from big tech vendors, a clear independent body to manage it, or simple tools that hide the complexity. Right now, none of those exist.

Is It Worth Exploring?

MCP is a good candidate for businesses that already rely on multiple systems and regularly integrate data for analysis or reporting. If your team regularly spends time exporting files, cross-referencing dashboards or transforming data for AI tools, MCP could reduce that effort and open the door to more brilliant, more contextual automation.

However, it is not a drop-in solution. It requires technical capability, patience, and a willingness to engage with early-stage tooling. If your business needs are simple or your existing BI tools already deliver value, you may find the cost of adoption outweighs the benefit, for now.

What’s Coming Next

MCP signals a shift towards assistants that can understand and operate within your business context. If adoption grows, we expect setup to become easier, integration coverage to expand, and security features to mature.

But that trajectory isn’t guaranteed. Without stronger vendor support, better tooling or clearer governance, MCP may remain a niche protocol for technically capable teams. The risk is not that MCP is flawed, but that it never crosses the threshold from experiment to infrastructure.

The technology has matured considerably since its November 2024 launch. Whilst still evolving, MCP now has backing from major tech companies, thousands of community-built servers, and official SDKs in multiple programming languages. However, security concerns and setup complexity remain barriers to widespread enterprise adoption.

Final Thought

AI becomes truly useful when it can operate within your context, not just discuss someone else’s. MCP is one possible step toward that future. It is unlikely to completely replace data dashboards or analysts anytime soon, but it could, over time, help answer the kinds of questions you rely on them for, if the infrastructure matures.

Whether MCP succeeds at scale is still an open question. Much depends on community momentum, vendor support, and whether practical tooling can close the gap between prototype and production.

It’s early, but it’s worth committing time to explore and understand, especially if you want to be ready for what might come next.

We’ve been experimenting with MCP and building practical AI integrations for clients. If you’re exploring how to connect AI with your business systems, we can help. Our Digital Product & AI service.

Further Reading

• Should We Optimise Content for AI? A Look at Generative Engine Optimisation (GEO)
• Designing for Everyone. What the European Accessibility Act Means for Your Digital Products.
• Sustainability Frameworks, Guidelines, Standards, and Protocols.
• Developing a Circular Business Model.
• Progress