Building an MCP Server for Your Business: A Step-by-Step Plan

An MCP server offers a unique opportunity for companies to connect AI assistants like Claude to internal systems, allowing users to easily and efficiently access valuable data. AI Insider developed its own MCP server for their data warehouse, with the aim of enabling employees to ask questions about numbers without requiring SQL knowledge or assistance from others. The project was completed within one workday, with two hours spent on the connection and five hours on explaining the data and optimizing answers, resulting in a significant improvement in accuracy.

The MCP server utilizes open standards, allowing AI assistants to independently request information. In this case, AI Insider chose not to use existing solutions like Supabase or Postgres, but instead built their own connector, so that every employee can ask questions without relying on individuals with access to the database. This encourages a culture of self-service within teams, greatly enhancing the speed and efficiency of data analysis.

The process involved several key steps, starting with ensuring security by implementing read-only SQL rights, followed by measuring the effectiveness of Claude's answers to predefined test questions. Documentation proved crucial, as it forms the basis of information provision to users. By adding detailed descriptions and explanations to the database, the team improved Claude's answer accuracy from 64% to 100%, while also reducing the number of tool calls per question.

AI Insider concludes that most time and effort goes into documentation and explanation, rather than in the technical development of the connection itself. For the maintenance of the MCP server, continuous monitoring of documentation and data is essential, especially with changes in the database, to ensure that information remains current and correct. This article not only provides insights into building an MCP server but also practical tips and a checklist for organizations considering implementing a similar solution.

Read the full article from AI Insider.