The Model Context Protocol (MCP) is an open standard, introduced by Anthropic, that lets an AI application like Claude connect to external tools and data sources through a single, consistent interface — instead of every integration needing its own custom, one-off plumbing.
Before USB-C, every device needed its own cable and its own adapter. MCP solves the same category of problem for AI tools: before it, every AI app that wanted to talk to, say, GitHub or a database needed a bespoke integration written specifically for that app. An MCP server exposes its capabilities once, in a standard shape, and any MCP-compatible client — Claude Desktop, Claude Code, or a growing list of others — can use it immediately.
A server built for MCP typically offers three kinds of things:
Most of the MCP servers you'll encounter day to day — for databases, project trackers, or something like a reminders app — mainly use tools: a small, well-typed set of actions the model can invoke on your behalf, with your explicit permission.
A regular API is built for an application developer to call, with documentation the developer reads once. MCP tools are built for a language model to call, with descriptions the model reads at runtime to decide when and how to use them. That distinction changes the design: tool descriptions need to be unambiguous enough for a model to use correctly without a human in the loop reading the docs first.
Early MCP servers mostly ran locally — a small process on your machine, launched via a command like npx, talking to Claude Desktop over stdio. That's still common for anything touching local files or dev tools. Increasingly, though, MCP servers run remotely over HTTP, so you can connect to a hosted service (say, a project tracker or a reminders app) via OAuth without installing anything locally at all.
If you want to see it in action with something concrete rather than abstract, connecting a reminders server to Claude Desktop is a good first MCP integration — it's a small, well-scoped tool surface (create, list, snooze, complete), and you'll feel the difference immediately: a reminder set mid-conversation that actually shows up in your inbox later, instead of evaporating when the chat ends. See our guide on setting up reminders in Claude Desktop for the two-minute walkthrough.