Model Context Protocol

Protocol tools published

Also known as: MCP, Anthropic MCP

Definition

An open protocol that standardizes how AI applications connect to external data sources and tools. MCP defines a client-server architecture where AI applications (clients) connect to MCP servers that expose tools, resources, and prompts through a unified interface. It aims to solve the "N×M integration problem"—instead of every app integrating with every tool, apps speak MCP and tools expose MCP servers.

What this is NOT

  • Not the same as function calling (MCP is a protocol; function calling is an API feature)
  • Not an Anthropic-only feature (MCP is open, works with any LLM)
  • Not a model (MCP is infrastructure, not an AI model)

Alternative Interpretations

Different communities use this term differently:

llm-practitioners

A specification (created by Anthropic, open source) for building integrations that LLM applications can discover and use. MCP servers expose capabilities; MCP clients (like Claude Desktop) connect to them.

Sources: Anthropic MCP documentation (modelcontextprotocol.io), MCP GitHub specification

Examples

  • Claude Desktop connecting to a GitHub MCP server to read/write repos
  • An MCP server exposing a company's internal knowledge base
  • Filesystem MCP server giving Claude access to local files
  • Database MCP server enabling natural language queries

Counterexamples

Things that might seem like Model Context Protocol but are not:

  • OpenAI's function calling API (that's a feature, not a protocol)
  • Direct API integration without MCP abstraction
  • A custom LangChain tool (tool, not protocol)

Relations

  • overlapsWith function-calling (MCP tools are invoked via function calling)
  • overlapsWith tool-definition (MCP standardizes tool definitions)
  • overlapsWith tool-use (MCP enables standardized tool use)

See Also

Implementations

Tools and frameworks that implement this concept: