Connect AI to the capabilities of your business through MCPs
Model Context Protocol (MCP) has quietly become the standard way to connect AI to the systems where your data lives. In this blog we cover off what it is, why it's taken off so quickly, where the hype outruns reality, and what it could mean for a Microsoft Azure or Fabric data estate.
Where the value in AI actually comes from
Most organisations have tried AI by now. Some have found the results underwhelming, though the limitation is rarely because the models aren't capable. Many others are seeing the business value from investment in AI and are yet to realise how far they could go with the models they are using.
An AI model can only work, learn, and deliver value with what it can see. In most businesses the data that matters sits in separate systems the AI can't reach.
Think about the last time your boss asked for a specific figure. You dug through your files, found the right report, scrolled to the right page, worked out the context behind the number, then wrote it all up and sent it back. That’s twenty minutes gone, maybe more.
Now picture asking the question in plain English and getting the figure, the context and a chart back in seconds, pulled straight from your own systems.
That’s what MCP makes possible.
When sliced bread first went on sale in 1928, nobody had invented a new food. It was the same loaf. What changed was how easy it was to use, and that small change is why the phrase has stuck for nearly a century.
MCP is the same kind of breakthrough. It doesn't make AI models smarter as such, and it doesn't create new data. What it does is make the data, systems and tools you already have become usable by AI, consistently, securely, and without a bespoke integration for every combination. AI models are increasingly able to reason and provide information, and if you can connect to your own data and systems they become immensely valuable.
That's why we think it's the best source of AI value available right now.
As an OpenAI Select Partner and Claude Select Services Partner we'll be upfront: we build MCP solutions for clients, so of course we're enthusiastic. That's exactly why we want to be honest about it. In this blog we cover what MCP actually is, the problem it solves, why the industry has rallied behind it, and the caveats you should know before you plug anything in.
What MCP actually is, in plain English
MCP is an open standard that lets AI applications connect to data sources and tools in one consistent way. It's often described as "USB-C for AI": one plug shape that works with any device that supports it.
There are three moving parts (Model Context Protocol, via Wikipedia):
The host is the AI application people use, such as Claude, ChatGPT, Microsoft Copilot Studio or an AI-enabled code editor.
The client sits inside the host and manages the connection to each MCP server.
The server wraps a system, such as a database, a document library or a CRM, and describes what it can do in language an AI model understands.
When someone asks a question, the flow is simple:
The client asks each server what tools and data it offers.
The AI model reads those descriptions and decides which tool will help.
The host asks the right server to run that tool.
The server does the work, such as running a query, and returns the result.
The AI uses that result to answer the person, grounded in real data.
The important bit: the AI model never needs to know how your systems work internally. It only needs to know what each server says it can do.
The tech problem MCP solves: one integration for every pair
Before MCP, every AI tool needed its own custom connector to every system it touched. Anthropic called this the "N×M" problem: N AI applications multiplied by M data sources equals a lot of integrations to build, secure and maintain (Wikipedia).
In practice, that looked familiar to most IT teams. One connector for the chatbot to read SharePoint. Another for the coding assistant to query the database. A third, slightly different one when the business switched AI vendors. Earlier approaches such as vendor-specific function calling and plug-in frameworks helped, but each locked you into one provider's way of doing things.
MCP turns N×M into N+M. Each system is exposed once, as an MCP server. Each AI application speaks MCP once, as a client. After that, any compliant client can work with any compliant server. Build the connection to your data estate once, and it works with the AI tools you use today and the ones you haven't chosen yet.
Why MCP has taken off so quickly
In under two years, MCP went from one company's open-source release to an industry standard backed by competitors. That's rare in technology, and it's the strongest signal that MCP is more than a passing trend.
The milestones tell the story:
November 2024: Anthropic open-sources MCP as a standard for connecting AI assistants to content repositories, business tools and development environments (Anthropic).
March–April 2025: OpenAI adopts MCP across its products (TechCrunch), and Google follows (TechCrunch).
December 2025: Anthropic donates MCP to the Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI and supported by Google, Microsoft and AWS. No single vendor owns it (Anthropic). At that point MCP already had more than 97 million monthly SDK downloads and 10,000 active servers (MCP blog).
July 2026: Maintainers finalise a major revision that makes the protocol stateless, simplifying how it scales (The Register).
Microsoft has been one of the most active adopters. The official C# SDK is maintained in collaboration with Microsoft (Microsoft Developer Blog). Copilot Studio can connect agents to MCP servers, with actions and knowledge added automatically and kept up to date (Microsoft). And Microsoft publishes MCP servers for services including SQL Server, Real-Time Intelligence and Microsoft Fabric itself (Microsoft Fabric blog). The practical upshot: when you invest in MCP, you're not betting on one AI vendor. You're investing in the plumbing that every major vendor now uses.
The honest part: what MCP isn't
Even sliced bread goes stale. MCP is a powerful standard, and a standard is only as good as how you implement it. Three caveats are worth knowing before you start.
MCP is plumbing, not intelligence. It gives an AI model access to your data and tools. It doesn't make your data clean, well-structured or well governed. If your data estate is messy, MCP will give AI fast access to the mess. Data readiness still comes first; our Data Solution Assessment can help you get there.
Security needs deliberate design. Security researchers have flagged real risks, including prompt injection and "poisoned" tools that could leak data through other connected tools (Wikipedia). The answer isn't to avoid MCP. It's to design for least-privilege access, authenticate properly, and control which servers your AI tools can reach.
The standard is still evolving. The July 2026 revision wasn't fully backward compatible, so newer servers may need a compatibility layer to work with older clients. That's a sign of a maturing standard, and it means MCP solutions need maintaining, not just building.
None of these are reasons to wait. They are reasons to build MCP with people who know where the edges are.
What MCP delivers from a Microsoft data estate
For organisations on Microsoft Azure or Microsoft Fabric data estates, MCP is where AI investment and data readiness finally meet. Most leaders we talk to have spent years building a data platform. The question now is whether their people can actually get answers from it.
Try a simple test. Could someone in your business ask, in plain English, for any of these and get a charted answer in minutes?
Market pricing profiles across your product range.
Sales volumes by geography this quarter versus last.
Your client base broken down by demographic.
If the honest answer is "we'd need to raise a ticket with the data team" or “let me come back to you on that”, that's the gap MCP closes. A well-built MCP server sits over your data in a Fabric or Azure platform, describes what's available and what it means, and lets AI tools query it safely on a user's behalf.
The benefits fall into two groups:
Faster, better-informed decisions. You get fully connected to your data, surface insights quickly, and can back decisions with evidence rather than gut feel.
Less time on recurring admin. Hand repeatable processes, such as setting up a new customer across several systems, to an AI agent working through MCP. Your team spends that time with the customer instead.
The off-the-shelf MCP servers are a great start, but the real value comes from enriched MCP servers built around your data, your business terms and your workflows. That's also where AI capability grows inside your team: people learn to make more advanced requests once the AI genuinely understands their data. As we found when measuring our own developers' AI use, the biggest gains come when AI stops feeling like a separate tool and becomes part of how people work.
MCP: Key Takeaways
Takeaway 1: MCP is an open standard for connecting AI to data and tools. Expose a system once, and any compliant AI tool can use it.
Takeaway 2: It replaces N×M custom integrations with N+M standard ones. Less to build, less to maintain, and no lock-in to a single AI vendor.
Takeaway 3: The industry has backed it. Anthropic, OpenAI, Google and Microsoft all support MCP, and it's now governed under the Linux Foundation.
Takeaway 4: It's plumbing, not magic. Data readiness, security design and ongoing maintenance still decide whether MCP delivers value.
Takeaway 5: The value is in enrichment. Generic servers get you started; tailored MCP servers deploy the reasoning AI models have developed onto your own data and workflows. This is where the significant business gains are.
MCP FAQs
Q: Is MCP a Microsoft product?
No. MCP was created by Anthropic and is now an open standard under the Linux Foundation. Microsoft supports it across Copilot Studio, Azure and Fabric, which makes it a natural fit for Microsoft-based organisations.
Q: Do we need Microsoft Fabric to use MCP?
No, MCP works with almost any data management system. If your data already lives in Fabric or Azure, you're well placed: Microsoft provides MCP servers for its own services, and enriched servers can be built on top of your data estate.
Q: Why would I need to develop custom MCPs if generic ones come with the product?
Pre-made generic MCPs give a wide range of access to data and leave it up to you, the client, to navigate and make decisions around which bits of data make sense and guess at your company’s interpretation of that. A custom MCP with data definitions and instructions built in makes a highly effective MCP tuned to your business, your data, and how you express what you want to know.
Q: Is MCP secure?
MCP can be implemented securely, but security isn't automatic. Design for least-privilege access, strong authentication and a controlled list of approved servers. Treat each MCP server like any other integration that touches sensitive data.
Q: How is MCP different from an API?
An API is built for developers to call from code. An MCP server describes its capabilities in a way an AI model can understand and use on a person's behalf. Many MCP servers sit on top of existing APIs, so your API investment isn't wasted.
Q: Do I need MCP to use AI in my business?
No. Tools like Microsoft Copilot, ChatGPT and Claude already deliver value on their own for drafting, summarising, research and analysing documents you give them. MCP becomes important when you want AI to work with your own data and systems: answering questions from your sales figures, looking up customer records, or completing tasks across several applications. A simple rule of thumb: if the answer you need lives inside your business systems, MCP is how AI gets to it.
Q: Where should I start?
Pick one high-value question your business struggles to answer today, and the data behind it. Prove MCP on that, measure the outcome honestly, then expand.
See MCP in action in 30 minutes
Reading about MCP only goes so far. Seeing an AI answer a real business question from live data, and chart it, is when it clicks.
Our complimentary Model Context Protocol (MCP): 30-minute introduction and demonstration available via the Microsoft Marketplace is built for CTOs, CIOs, CDOs, CMOs and Heads of Data and Analytics. In 30 minutes, we'll introduce our approach to developing enriched MCP servers that integrate closely with your data and AI tools and show how they can work in your own data estate.
MCP may not be better than sliced bread. But for getting real value from your data, it's the closest thing we've seen.
Book your free MCP introduction or talk to our team.
Sources
Model Context Protocol (MCP): 30-minute introduction and demonstration, Microsoft Marketplace
Model Context Protocol, Wikipedia
Introducing the Model Context Protocol, Anthropic
Introducing MCP in Copilot Studio, Microsoft
Start with one question
Tell us the business question you struggle to answer today. We’ll show you how MCP could answer it from your own data, and what it would take to get there.
