Data trapped in isolated systems has frustrated people for as long as humans have tried to store and share knowledge. Long before software, knowledge was often locked inside specific cities, libraries, institutions, and political systems.
Today, the problem remains. Valuable information exists, but it’s difficult to access, connect, and use across different systems. That leads to challenges like:
- Reduced efficiency, as people waste time searching for information and asking for information in meetings, instead of simply retrieving it.
- Impaired decision-making, as decisions are based on siloed data.
- Complicated automation, when workflows depend on fragile integrations, manual workarounds or incomplete data.
What are data silos?
According to IBM, data silos are “isolated collections of data that prevent data sharing between different departments, systems and business units.“
In an organisation, departments often use their own tools, which makes it difficult to share data directly with other teams without time-intensive or error-prone workarounds.
Here’s a real-world example.
Data silo example: marketing, sales, and support
A company might use HubSpot for marketing campaigns, Salesforce for sales pipelines, and Zendesk for customer support.
Each system contains valuable customer information, but that information remains siloed.
Marketing sees campaign engagement, the sales team sees deal status and revenue, support sees complaints and tickets, but none of them has the full picture.
That can create real problems. Sales may contact a customer without knowing they recently opened several support tickets. Marketing may send promotional emails to users who are already frustrated. Support may miss important context about a customer’s account value or recent sales conversations.
Why AI is making the problem more visible
Data silos are not new. So why are they getting so much attention now? One reason is AI.
Before generative AI became part of everyday work, teams often used manual workarounds to cope with data silos. Departments worked in separate platforms, and when needed, people copied information manually between systems.
There were integrations, of course. But often, they were usually built for one specific purpose and required ongoing maintenance. Of course, this setup was inefficient, but teams learned to work around it.
Modern AI tools are increasingly expected to work across many different systems. A single agent might need to:
- Pull customer details from a CRM
- Look at support tickets
- Check analytics
- Update tasks in a project management tool
- Trigger actions in another platform entirely
To do that well, systems need to share information smoothly with each other. But most tools weren’t built to work together this seamlessly. They use different data structures, APIs behave differently, and many integrations are custom-built for one narrow use case.
As a result, familiar problems become more visible. Information is fragmented, systems don’t communicate properly, and sometimes, integrations just break. As a result, important context gets lost between tools.
How companies usually deal with data silos
In addition to the workarounds mentioned above, organisations are continuously trying to address data silos in several ways. Some centralise data in warehouses or lakes. Others build API integrations, use automation platforms, or rely on business intelligence tools to bring information together. These traditional approaches to data silos help organisations collect, structure, and analyse data from different sources. Their main goal is to bring information together.
That works for many reporting and analytics use cases, but it becomes less flexible when systems need to interact dynamically. This is especially true for AI-powered tools. An AI assistant may not just need to read data from one place. It may need to understand what tools are available, access the right context, and trigger the right action across different systems.
That is where the conversation shifts from simply “breaking down data silos” to improving interoperability between tools.
From data centralisation to tool interoperability
All of these traditional approaches are still important, especially for reporting, analytics, and creating a reliable view of business data. But increasingly often, systems don’t just need to send data from one place to another. They also need to expose what they can do, what information they hold, and how other tools can interact with them. That’s where MCP becomes relevant.
What’s MCP?
MCP, short for Model Context Protocol, was introduced by Anthropic in 2024 as an open standard that helps AI tools work across existing business systems without every connection having to be built from scratch.
Instead of building custom integrations for every possible workflow, MCP creates a consistent way for AI to access external tools, data sources, and workflows. In simple terms, it acts like a common language between AI systems and the tools they interact with.
Rather than forcing organisations to centralise all their data into one system, MCP can help AI applications access data and tools where they already live.
If a company wants an AI assistant to work with a web analytics platform, CRM, help desk, or documentation system, someone needs to provide or build an MCP server for that system.
That means MCP isn’t universally available across all software yet. It can be implemented for many applications, but only systems with MCP support or an MCP server can participate.
Who offers MCP support or runs MCP servers?
There are three common ways of making MCP usable:
- A software company, such as a CRM, analytics platform, help desk, or project management tool, can build and offer an official MCP server for its own product.
- A company can build its own MCP server for internal systems.
- A vendor, agency, consultant, or open-source maintainer can build MCP servers for popular tools.
A simple MCP example
Imagine a marketing team notices that conversions from paid campaigns dropped sharply last week. To investigate, they need to understand whether the issue came from campaign performance, website behaviour, tracking changes, or something else. An AI assistant could help with that investigation, but only if it can access the right context across different systems:
- Web analytics data showing traffic, conversion rates, referrers, and campaign performance
- Advertising data showing spend, clicks, and campaign changes
- CRM data showing lead quality or sales outcomes
- Support tickets mentioning checkout issues or broken forms
- Internal release notes showing recent website or tracking changes
Without a shared way to access that context, teams often fall back on manual checks, exports, dashboards, or one-off integrations.
For tools and data sources that support MCP, the protocol offers a more consistent way to expose relevant context and actions to an AI application. According to the official MCP documentation, it’s “an open-source standard for connecting AI applications to external systems“, such as data sources, tools and workflows.
This way, instead of treating every system as a separate integration project, an AI application could use a common protocol to discover what information is available, retrieve the right context, and support the team’s investigation. The systems remain separate, but the way the AI application works across those systems becomes more consistent.
That’s especially important when scaling: Every custom integration adds maintenance overhead and becomes more difficult to maintain as the number of systems in use grows and the need for interoperability increases.
What MCP isn’t
MCP isn’t a replacement for APIs, databases, or data warehouses. It also doesn’t magically eliminate data silos. The underlying systems still exist separately, and organisations still need governance, permissions, and reliable infrastructure.
What MCP does is reduce some of the friction caused by disconnected systems. It doesn’t replace APIs. In many cases, it uses them behind the scenes. APIs remain the way individual systems expose data and actions. MCP provides a common layer that helps AI applications discover and use those capabilities more consistently across different tools. In practice, MCP can help businesses make their existing tools more usable for AI-powered workflows, while still keeping data in the systems where it already lives.
Access still needs to be controlled
Making tools easier to connect doesn’t mean everything should be accessible to everyone.
If an AI application can operate across CRM data, support tickets, analytics, and internal documentation, access control becomes even more important. Organisations need clear rules around permissions, data governance, authentication, and auditability.
While MCP helps standardise how tools and context are exposed, it doesn’t remove the need for responsible implementation.
Why interoperability matters now
Most companies rely on dozens of tools that weren’t made to work well together. As automation and AI-powered workflows become more common, those gaps become harder to ignore. That’s why interoperability is such an important topic right now — with protocols like MCP as a way to expose information, context, and actions more consistently, without relying on endless custom integrations.
The direction is clear: the easier tools can communicate with each other, the more useful they become.
Want analytics that fits into your stack without giving up data ownership? Start your 21-day free trial.
