MCP (Model Context Protocol)

MCP (Model Context Protocol) is an open standard that defines how AI applications expose tools, data, and prompts to LLMs. Think of it as USB-C for AI: instead of every chatbot reinventing the integration with GitHub, Notion, your filesystem, and your database, you write one MCP server and any MCP-aware client (Claude Desktop, Cursor, ChatGPT, Zed, Continue, custom apps) can use it. The protocol has three primitives: tools (functions the model can call), resources (data the model can read, like files or DB rows), and prompts (reusable templates). A server advertises what it offers, a client connects, and the LLM gets a typed list of capabilities it can invoke. Messages are JSON-RPC 2.0, sent over stdio for a local server or Streamable HTTP for a remote one. Anthropic open-sourced MCP on November 25, 2024. OpenAI added it to its Agents SDK in March 2025 and Google to the Gemini API SDK by May 2025. On December 9, 2025, Anthropic donated the protocol to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block, and OpenAI, so no single lab owns it. The spec keeps moving. The 2026-07-28 revision made the core stateless: the initialize handshake and the Mcp-Session-Id header are gone, and servers answer a required server/discover call instead. The same revision moved Tasks out of the core into an official extension (MCP Apps is another) and deprecated Roots, Sampling, and Logging, which cannot be removed before the first spec revision released on or after July 28, 2027. With thousands of MCP servers already available for everything from Stripe and Linear to Postgres and Kubernetes, building tool integrations without MCP means rebuilding something that is already standardized. The catch: MCP standardizes the wire format, not the quality. A bad MCP server with vague tool descriptions and leaky abstractions makes any model worse. The good ones look like well-designed APIs — narrow tools, clear naming, structured errors.