Focus_mcp_sql

What is Focus_mcp_sql?

Focus_mcp_sql is a Model Context Protocol (MCP) server that enables AI assistants to convert natural language into SQL statements. It employs a transparent two-step generation process (keywords first, then SQL) to reduce LLM hallucinations and build trust in the generated SQL results.

How to use Focus_mcp_sql?

Clone the repository, build the server with Gradle (requires JDK 23+ and Gradle 8.12+), then add the server to your MCP settings file with the java -jar command. After configuration, use the gptText2sqlStart tool to initialize the model with table schema (and a Datafocus bearer token), then use gptText2sqlChat to convert natural language queries into SQL.

Key features of Focus_mcp_sql

• Transparent two-step generation: keywords then SQL
• Low hallucination risk through keyword verification
• Fast, deterministic keyword-to-SQL conversion
• Low cost by reducing reliance on large models
• Non-technical user friendly with easy keyword checking
• Requires only a Datafocus bearer token for authentication

Use cases of Focus_mcp_sql

• Non-technical users querying databases via natural language
• Scenarios where SQL accuracy and transparency are critical
• Reducing operational costs by minimizing LLM inference usage
• Providing a verifiable SQL generation pipeline for business analysts

FAQ from Focus_mcp_sql

How does Focus_mcp_sql reduce hallucinations?

It uses a two-step approach: first generates keywords from the natural language query, then maps those keywords deterministically to SQL. This keyword verification step makes the process transparent and controllable, lowering hallucination risk.

What are the prerequisites for running Focus_mcp_sql?

You need JDK 23 or higher, Gradle 8.12 or higher, and a registered Datafocus account to obtain a bearer token for authentication.

Where does my data live during SQL generation?

The table schema and bearer token are sent to the server as parameters in the gptText2sqlStart tool. The actual database is not accessed directly; only the metadata provided by the user is used.

What transport does Focus_mcp_sql use?

The server communicates over stdio via the MCP protocol, as configured in the MCP settings file with a java -jar command.

What limits should I be aware of?

The server relies on the Datafocus API for token authentication and keyword generation. It is designed for structured table schemas and does not handle unstructured data or complex joins without explicit relation definitions.