MCP (Model Context Protocol)

MCP (Model Context Protocol)

Also called: 模型上下文协议 · Model Context Protocol

An open standard from Anthropic (Nov 2024) that lets agents discover and call external tools, data, and prompts in a uniform way — often called "USB-C for AI apps."

Host + Client(LLM app)Servertools /resourcescallresult
Schematic (simplified)

What it solves

Before MCP, every data source or tool needed a custom integration written for a specific model and app — an M×N glue-code problem. MCP defines a standard interface layer: a tool provider implements an MCP Server once, and any MCP-capable host application can use it plug-and-play.

Technical shape

MCP is built on JSON-RPC 2.0 with a client–server structure. A server can expose three primitives: Tools (functions the model can call), Resources (data it can read), and Prompts (pre-defined templates). Transports include local stdio and remote Streamable HTTP. The current stable spec is 2025-11-25 (with a 2026-07-28 release candidate); governance is now hosted by the Agentic AI Foundation.

Common misconceptions

  • MCP does not replace function calling — it standardises the tool *supply side*; the model still uses tools via tool calls.
  • MCP is for "agent ↔ tools/data"; "agent ↔ agent" collaboration is the domain of protocols like A2A.

Related terms

Sources

  1. Model Context Protocol — Specification
  2. Anthropic — Introducing the Model Context Protocol (Nov 2024)
  3. MCP Blog — One Year of MCP (Nov 2025 spec release)

Compiled 2026-08-29 · This glossary is compiled from public papers, official specifications, and common industry definitions, and is updated as the field evolves. Corrections welcome.