MCP server for data analysis

Overview

The company collected data from several external APIs and combined it into a complex relational dataset. The data was useful, but answering questions about it required knowledge of both SQL and the underlying schema.

FerrisDev built an MCP server that gave an LLM access to the dataset. Users could ask questions in natural language and receive plain-language answers together with the relevant tabular results.

Why the Data Was Hard to Access

The dataset contained enough tables and relationships that users could not explore it effectively without technical knowledge. Even a straightforward business question first had to be translated into the correct joins, filters, and aggregations.

The goal was to make that information accessible without requiring each user to learn the schema or write SQL.

From Natural Language to SQL

Data collected from multiple APIs was structured and loaded into a relational dataset. An MCP server exposed that dataset to an LLM, which translated a user's question into SQL and executed the query through the server.

Presenting the Results

The LLM then interpreted the query results and returned both an explanation and a table. Users could ask follow-up questions in the same interface instead of navigating the database directly.

Outcome

Users could investigate the dataset and obtain answers in a form they could readily understand, without having to write SQL or study the underlying schema.

Where an MCP Data Interface Fits

This approach is useful when an organization has a valuable but complex SQL dataset that remains difficult for non-technical users to access. An MCP interface can connect the dataset to an LLM and turn natural-language questions into answers supported by query results.