Modular acts as an intelligent proxy for Model Context Protocol (MCP) servers, addressing the challenge of managing large tool collections for Language Model (LLM) applications. It efficiently groups tools from various upstream MCP servers, exposing them through a single endpoint. By providing group descriptions to the LLM and loading detailed tool schemas only when requested, Modular drastically reduces context overload and streamlines the LLM's tool discovery and execution process. This architecture ensures LLMs can access a vast array of functionalities without being overwhelmed by unnecessary information.