Abstract

As large language models (LLMs) demonstrate increasingly powerful reasoning and orchestration capabilities, LLM-based agents are rapidly adopted for complex data-related tasks. Despite this progress, the current design of how LLMs interact with databases exhibits critical limitations in usability, security, privilege management, and data transmission efficiency. To address these challenges, we introduce BridgeScope, a universal toolkit that bridges LLMs and databases through three key innovations. First, it modularizes SQL operations into fine-grained tools for context retrieval, CRUD execution, and ACID-compliant transaction management. This design enables more precise, LLM-friendly controls over database functionality. Second, it aligns tool implementations with database privileges and user-defined security policies to steer LLMs away from unsafe or unauthorized operations, which not only safeguards database security but also enhances task execution efficiency by enabling early identification and termination of infeasible tasks. Third, it introduces a proxy mechanism that supports seamless data transfer between tools, thereby bypassing the transmission bottlenecks via LLMs. All of these designs are database-agnostic and can be transparently integrated with existing agent architectures. We also release an open-source implementation of BridgeScope for PostgreSQL. Evaluations on two novel benchmarks demonstrate that BridgeScope enables LLM agents to interact with databases more effectively. It reduces token usage by up to 80% through improved security awareness and uniquely supports data-intensive workflows beyond existing toolkits. These results establish BridgeScope as a robust foundation for next-generation intelligent data automation.