Model Context Protocol (MCP)

The Model Context Protocol (MCP) is "an open protocol that enables seamless integration between LLM applications and external data sources and tools". It uses JSON-RPC 2.0 messages between hosts (the AI applications), the clients inside them and servers that offer tools, resources and prompts.¹ Anthropic open-sourced it on November 25, 2024, as "a new standard for connecting AI assistants to the systems where data lives".²

MCP names each version by date: a revision's YYYY-MM-DD is the last date backwards-incompatible changes were made, and the current version is 2026-07-28.³ That revision made the protocol stateless, removing the initialize handshake, and added a mandatory server/discover call through which a server advertises its protocol versions, capabilities and identity.⁴ OAuth-based authorization is optional; a server that uses it acts as an OAuth 2.1 resource server.⁵ How the Model Context Protocol Works (2026-07-28 Revision) walks through the roles, transports and revisions.

On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI.⁶⁷ Anthropic stated then that there were more than 10,000 active public MCP servers and 97M+ monthly SDK downloads across Python and TypeScript Vendor-stated.⁶ The TypeScript SDK alone was downloaded 261,880,885 times from npm in the 30 days to October 16, 2026 Measured, and the official MCP Registry listed 42,320 active servers when MetaPrompt counted it on October 13, 2026 Measured.⁸⁹ MCP's own authorization identifies the client application and its user to a tool server rather than giving an agent an identity of its own; AI Agent Identity and Trust Standards sets it beside the other agent identity specifications.⁵ The A2A specification describes MCP and A2A as complementary: MCP for connecting agents to tools and data, A2A for agents working with each other.¹⁰