MCP

MCP is an open protocol that lets any AI app discover and use tools, files, and prompts from external systems through a shared client-server language.

MCP exists to kill the N×M connector problem — without a shared protocol, every AI app rebuilds the same GitHub, Slack, and database integrations from scratch.

MCP has a host that contains clients, each talking to one server over stdio (local) or Streamable HTTP (remote) — and since 2026 the HTTP core is stateless.

MCP servers expose three primitives — tools that act, resources that are read into context, and prompts that are reusable starting recipes.

Remote MCP uses OAuth 2.1 with audience-bound tokens; local stdio servers use env credentials — never pass tokens through.

An MCP server wraps one system behind clear tools, resources, and prompts, served over stdio or HTTP with an official SDK.

A host consumes MCP by connecting a client per server, mapping the catalog into function calling, and enforcing allow-lists and approvals.

MCP is the plug; agents, IDEs, and products are hosts that decide when to call — a product can also publish a server so other hosts reach it.

MCP's real risks are poisoned tool descriptions, confused-deputy OAuth, token passthrough, and untrusted servers — treat catalogs as untrusted.