MCP (Model Context Protocol)

MCP is an open protocol that gives AI applications one standard way to reach outside systems. A tool vendor publishes an MCP server that describes what can be done with its data. An AI application that supports MCP connects to that server, reads the description and lets the model call the tools it finds there. The same server works in every compatible assistant, so a vendor does not need one integration per AI product.

The protocol defines three kinds of things a server can offer. Tools are actions, such as running a report or creating a payment link. Resources are data the assistant can read. Prompts are reusable instruction templates. Messages between client and server use JSON-RPC, carried either over a local process (stdio) or over HTTPS (Streamable HTTP) for hosted servers.

For an online store, MCP is the mechanism behind phrases like 'connect Claude to Google Ads' or 'let a shopping assistant read your catalog'. It does not decide what the assistant may do: that is set by the server's tools and by the permissions you grant when you connect. Reading the directory listing for a server tells you both.