What is an MCP server?

MCP servers (Model Context Protocol servers) are standardized interfaces through which AI language models can access external data sources, tools, and services in a structured and secure way. Instead of building a separate integration for every combination of model and data source, they follow a unified protocol. With BI solutions from Bissantz, MCP servers enable natural-language queries that provide answers based on all relevant data sources – while the company retains full control over its data.

MCP servers (model context protocol servers) are standardized interfaces through which AI language models can access external data sources, tools, and services in a structured way.

MCP servers are neither AI models nor a replacement for a data warehouse. They are an infrastructure component – comparable to an API, but with the specific purpose of integrating language models into business processes in a secure and structured manner.

In controlling and business intelligence, MCP marks a turning point: for the first time, AI assistants can be systematically and reproducibly connected to business data – without proprietary point-to-point solutions for every combination of model and data source.

An MCP server provides AI language models with structured access to external data and functions. It receives requests from an AI assistant, mediates access to the configured resources, and returns the results in a format that the model can process directly.

The model context protocol defines:

• Which data is accessible – and which is not
• Which operations are permitted – e.g., queries, calculations, and filtering
• What does not leave the system – to protect sensitive company data

It is therefore less a data source than a controlled access layer between AI and company data.

MCP follows a client-server architecture. The MCP client is typically an AI assistant or an application that embeds language models. The MCP server provides defined resources: data, functions, and contextual information that the model is permitted to access during a request.

Communication between client and server takes place via a standardized protocol. This creates a clean separation: the language model does not need to know how a database is technically structured. It asks the server for information, and the server delivers it in a format that the model can understand and process.

MCP distinguishes between three types of server offerings:

• Resources: structured data that the model can read (e.g., tables, reports, master data)
• Tools: functions that the model can call (e.g., perform calculations, filter data, launch queries)
• Prompts: predefined query templates that standardize recurring tasks

Controlling depends on timeliness, reliability, and traceability. When MCP is used, AI assistants do not receive uncontrolled access to database layers. Instead, they communicate through clearly defined, purposefully designed interfaces.

An MCP server for controlling could, for example, provide the following resources:

• current actual figures from the data warehouse
• plan values and forecasts
• variance analyses at the push of a button
• comments and explanations from reporting

A controller working with an AI-supported controlling system such as DeltaMaster from Bissantz can ask the AI assistant a question in natural language. The assistant retrieves the relevant data via the MCP server, performs the calculation, and returns a substantiated answer. Data sovereignty remains with the company.

Before MCP, integrating language models into business systems was largely a bespoke development task. Every connection between an AI model and a database system required its own adapter, authentication logic, and error handling. The result was difficult-to-maintain point-to-point solutions that had to be rebuilt whenever the model changed.

MCP solves this problem through standardization. An MCP-compatible server can be connected to any MCP-enabled client – regardless of which language model is running behind it.

In a business context, the question of who is allowed to access which data is not a technical side issue. MCP servers can control access rights at a granular level. Which resources a model may see, which functions it may call, and which data must never leave the system – all of this can be configured and enforced on the server side.

MCP servers can therefore control access rights at a granular level – provided they are configured accordingly and integrated into the company’s existing security architecture. In this way, MCP can contribute to the auditability of AI in controlling: every request, every response, and every data access can be logged.

Bissantz uses MCP servers to provide powerful AI functions in DeltaMaster. Through the MCP protocol, all connected data sources can be made uniformly and easily accessible to Agentic BI. Companies do not need to develop complex individual integrations: one standardized interface is sufficient to give AI secure access to the data that is actually relevant for controlling and reporting.

The result: Users can ask questions in natural language – e.g. about variances, contribution margins, or forecast trends – and receive answers based on all relevant information sources. Data sovereignty remains fully with the company. What the AI may see, what it can calculate, and what does not leave the system is controlled by MCP – transparently, verifiably, and according to the rules defined by the company itself.

MCP servers solve one of the central challenges of using AI in companies: the controlled, standardized, and secure connection of language models to internal and external sources of information. In controlling, they enable queries based on a comprehensive data foundation – without data loss, without proprietary point-to-point solutions, and without dependence on a specific AI model. Bissantz integrates MCP into DeltaMaster and DeltaApp, making AI-supported controlling transparent, auditable, and future-proof.