MCP Guide

Enterprise AI systems can only deliver real value when they have proper access to data, tools, and workflows. Model Context Protocol (MCP) is an open standard that helps engineering teams connect AI applications to authorized enterprise systems in a standardized way.

MCP is not a model or a database itself, but a protocol positioned at the application and integration layer of the AI stack. This allows organizations to reduce the burden of individual integration work that arises when connecting AI applications to enterprise systems, and to easily extend and maintain systems as requirements change.

MCP operates through a client-server architecture consisting of a Host, Client, Server, and Transport mechanism.

• Host: The AI application or environment the user interacts with (e.g., an assistant, IDE, or internal AI tool)
• Client: A component within the host that maintains a connection with a specific MCP server
• Server: A program that exposes context or functionality from external systems such as databases, document stores, or SaaS applications
• Transport mechanism: The communication method between client and server (e.g., stdio for local connections, Streamable HTTP for remote connections)

MCP servers provide three main types of capabilities.

• Resources: Contextual information available for AI applications to use, such as files, records, and schemas
• Tools: Defined functions that AI can call, such as querying databases, calling APIs, or updating business systems
• Prompts: Reusable templates or structured instructions provided by the server for specific interactions or workflows

MCP is closely related to API, RAG, Function calling, and Agent, but each plays a different role. MCP does not replace APIs; rather, it standardizes how AI can discover and use functionality exposed through APIs, and provides a standard path for RAG to access knowledge bases. It also standardizes how tools are exposed through Function calling, making it easy for compatible applications to use them.