Overview
This page is the atomic definition. The MCP deep-dive lives at mcp-servers.
Definition
Model Context Protocol (MCP) is an open standard introduced by Anthropic in late 2024 that defines a transport-agnostic JSON-RPC interface between AI models (clients) and external capability providers (servers). An MCP server exposes three primitives: Tools (functions the model can invoke), Resources (data the model can read), and Prompts (reusable prompt templates). The model discovers available tools via a tools/list call, then invokes them via tools/call. Transports include stdio (local subprocess) and HTTP with Server-Sent Events (remote). MCP decouples tool implementation from model implementation: a single MCP server works with any client that speaks the protocol. Claude Code, Cursor, and several other AI editors ship as MCP clients. The protocol addresses the N*M integration problem by creating a single standard instead of per-model, per-tool adapters.
When it applies
Use MCP when building tools for AI agents that should be reusable across different clients or models. For single-model integrations where full portability is not needed, function-calling via the provider’s native API is simpler.
Example
An MCP server wraps a Postgres database. Claude Code (MCP client) calls tools/list, discovers run_sql, and uses it to query the schema. The same server works with any MCP-compatible client.
Related concepts
- mcp-servers - the full MCP setup guide including server and client configuration.
- tool-use - MCP tool calls are a standardized form of tool use.
- function-calling - the provider-specific API that MCP generalizes.
- multi-agent - MCP enables agent-to-agent tool discovery.
- claude-code - Claude Code is the reference MCP client.
Citing this term
See MCP (Model Context Protocol) (llmbestpractices.com/glossary/mcp).