記事の概要
The Model Context Protocol (MCP) is introduced as a standardized framework for Large Language Models (LLMs) to effectively manage and receive contextual information from external tools and systems.
- MCP aims to overcome the inherent limitations of fixed context windows by enabling proactive, structured injection of relevant data.
- The protocol defines 'MCP Servers' as providers of contextual information and 'MCP Clients' as the LLMs consuming this context, interacting via a `ModelContextProvider` interface.
- Benefits include significantly enhanced LLM capabilities, a reduction in model hallucinations, improved tool integration, and greater control over the contextual data presented to the model.
- MCP also proposes a 'Context Description Language (CDL)' to structure context and is positioned as an advanced layer building upon existing concepts like function calling and Retrieval Augmented Generation (RAG).