MCP Introduction

Model Context Protocol (MCP) is a standardized protocol designed to decouple large language model tool call from the tightly coupled Function Calling model. In the traditional approach, AI tool calls are bound to a specific system—for example, ChatGPT function calls can only access its own plugins. MCP separates the conversation layer from the tool layer: the conversation client only needs to implement MCP support, while MCP tools can be developed and maintained independently across different backends.

Starting from DolphinDB 3.00.4, DolphinDB provides a native MCP Server implementation that allows users to convert custom functions into MCP tools, enabling agents to leverage DolphinDB's data analysis capabilities.

• Tool definitions are centrally stored on a DolphinDB controller node that supports MCP Server. During execution, the function logic runs on the connected data node or compute node.
• The conversation interface (Agent frontend) can be any MCP-compatible client, such as Visual Studio Code Copilot, Claude, Cherry Studio, or the AI Assistant of DolphinDB Starfish.
• The large language model acts as the orchestration engine on the client side, determining which tool to call and what parameters to provide based on the conversation context.