Model Context Protocol (MCP) is a standardized protocol designed to provide Large Language Models (LLMs) with external context. It acts as an adaptation layer between AI applications and data sources, standardizing how context is retrieved and how tools are invoked.
Key components include:
- Resources: Standardized interfaces for static, read-only data retrieval, effectively normalizing the RAG (Retrieval-Augmented Generation) process.
- Tools: Standardized interfaces for function calling, allowing AI applications to execute specific logic via a consistent protocol regardless of the underlying model's native function-calling capabilities.
MCP uses jsonrpc 2.0 for message formatting and supports two primary transport mechanisms: stdio and httpstream.