Show HN: SQL MCP Server – 61.37% on DataAgentBench with GLM-5.2 (github.com)

DataBridge, an open-source MCP server, has been introduced, achieving a notable 61.37% success rate on the DataAgentBench benchmark when tested with the GLM-5.2 model. This innovative solution enables AI agents like Claude, GPT, and Gemini to interact efficiently with multiple heterogeneous databases, such as PostgreSQL and MongoDB. DataBridge significantly enhances the querying process by acting as an intermediary that not only facilitates connections and enforces safety but also automates schema learning, cross-database joins, and post-query transformations. This development is crucial for the AI/ML community, as it addresses common pitfalls faced by current AI agents, including silent errors in data retrieval, the absence of safety measures for destructive database commands, and inefficient data handling. By ensuring that queries are read-only by default and introducing an intelligent join registry that minimizes manual intervention, DataBridge simplifies complex querying across diverse systems, transforming the way AI applications access and analyze enterprise data. Its ability to perform tasks like schema caching and execute multi-DB plans without requiring intricate configurations makes it a powerful tool for enhancing data-driven decision-making in various industries.