Data & AI platform by Data Analytics
What is MCP and what does it mean for your organisation?
MCP stands for Model Context Protocol. It describes how an AI application can exchange context and capabilities with connected software. Think of it as an agreed way to communicate, not a new database or a smarter model. You do not need to read protocol messages for an initial assessment. It matters more to understand which parties are involved, what information leaves your organisation and who decides what may happen.
The examples on this page are hypothetical situations, not customer cases.
Meet our team
Trusted by data-driven organisations
01
Three roles in a simple conversation
The host is the application your employee uses to work with AI. Within that host, an MCP client handles the connection to an MCP server. That server exposes selected information or functions from underlying systems. The host can work with multiple servers, depending on support and configuration. Example: a hypothetical facilities assistant receives a question about which locations have overdue maintenance. A client contacts a server exposing permitted maintenance data. The language model can summarise the result. The maintenance system remains the source; the model does not suddenly become the owner or administrator of that record.
02
Tools, resources and prompts have different jobs
A tool is a callable function, such as retrieving a maintenance overview for a selected region. A resource supplies readable context, such as the meaning of maintenance categories. A prompt is a reusable instruction template that helps formulate a task. Not every server or host supports the same capabilities. Ask what is actually available in the intended setup. These terms are not quality labels: a tool can contain an incorrect calculation and a resource can be outdated. Appoint owners for content and definitions. Also check whether the user can see the source and reference date associated with a result before making a decision.
03
From your question to an answer
A common path begins with your question in the host. Based on available capabilities, the model may propose a tool call. The application handles any approval and the server checks the request before the underlying function runs. The result returns to the host and can provide context for a model response. The exact interaction varies by implementation. Example: a hypothetical employee requests an overview of inspections in their region. The server must restrict that region according to their permissions, not merely their wording. A convincing answer still needs checking, especially when data is missing or an inspection has not yet been recorded.
04
What MCP does not solve for you
MCP is not a model, an ETL process or a replacement for a data model. It does not automatically move all your sources into one reliable data foundation or fix duplicate locations and conflicting definitions. An API or integration process may still be needed to expose and organise source data. Nor does MCP determine where the language model runs. If the host passes data to an external model provider, it is processed externally there. Discuss beforehand which fields are necessary, which agreements apply and whether employees may send sensitive details. A local server does not automatically mean all subsequent processing stays local.
05
Permission is more than a polite instruction
An instruction such as “change nothing” is not a technical access restriction. Read-only access must be enforced by the server and permissions in the database or source system. Any changes require explicit permission and separate authority. Execution must check the current situation and business rules within the transaction. For inspection scheduling, consider checking for duplicate bookings before committing a change. Also agree what to record: tool name, time and status are often more useful than entire conversations. Avoid sensitive inputs and results in logs. A clear process helps employees understand when they are merely reading information and when they are actually starting an action.
06
Start with the data foundation for your question
ENABLE supports the role of a coherent data foundation with integration and dashboards. For an MCP application, those data and definitions must be usable for the chosen question. Connection through MCP depends on implementation; this explanation does not confirm any particular available ENABLE connector. Start with a small information task and determine what a correct answer must contain. Compare a trial result with your existing reporting and have a subject-matter owner assess differences. Only then discuss wider use. During preparation, ask about user permissions, source freshness, external model processing and operations. This keeps the assessment understandable without hiding technical choices behind a protocol’s name.
Frequently asked questions
Do you need to code to use MCP?
Not always as an end user. Your organisation must still have the chosen connection technically assessed and managed.
Can MCP make answers error-free?
No. Source quality, definitions, selection and model interpretation still matter; verify important answers.
Does AI gain access to all data?
That should not happen automatically. The setup should expose only necessary information and permissions.
Does this make a dashboard unnecessary?
No. A dashboard provides consistent comparisons; a conversation can support additional questions using the same controlled definitions.
Technical documentation
Discuss your question with Data Analytics
Which sources, users and checks does your application need? Discuss your question in an ENABLE demo.
Meet our team