Model Context Protocol (MCP) server features overview

This section provides an overview of the Model Context Protocol server features, including prompts, resources, tools, and utilities.

MCP server features

The MCP server provides three core primitives: Prompts for user-controlled templates, Resources for application-managed contextual data, and Tools for model-executed functions. These enable rich interactions between clients, servers, and language models with distinct control hierarchies.

MCP server features prompts

MCP prompts are user-controlled templates that guide LLM interactions. Servers expose prompts through list and get operations, supporting arguments, icons, and various content types including text, images, audio, and embedded resources. The protocol supports pagination and change notifications.

MCP server features resources

MCP resources let servers share contextual data like files and database schemas with clients. Each resource has a unique URI. The protocol supports subscriptions for real-time updates, resource templates for parameterized data, and annotations for metadata like audience and priority.

MCP server features tools

MCP tools enable language models to interact with external systems through callable functions. They include input and output schemas, support structured and unstructured content types, and require proper validation and access controls. Tool execution errors help models self-correct.

MCP server features utilities

MCP utilities include completion for contextual suggestions, logging for structured messages with severity levels, and pagination using cursors for efficient large dataset handling.