Model Context Protocol (MCP)

Model Context Protocol (MCP) is an open protocol for exposing tools, resources, and prompts from a server to a compatible AI client through defined transports and schemas.

An MCP server describes capabilities and handles requests from a client that implements the protocol. A server can expose database queries, file resources, or application operations, while each client decides which protocol features, transports, authentication, and approval flows it supports. Shared protocol messages can reduce client-specific integration code, but they do not make every server work unchanged in every product.

Builder example

MCP separates capability publication from a particular assistant UI. Reuse still depends on compatible protocol versions, schemas, transport, authentication, deployment, and client policy. Connecting a server exposes only the capabilities and credentials configured for that connection, not automatically everything the server or client can reach.

You want Claude, Cursor, and another agent to query the same project notes without rebuilding the integration three times.

Expose the notes through a Model Context Protocol (MCP) server, then keep permissions narrow and version the server contract.

Common confusion: MCP handles the connection, not the security. The tool descriptions and data that flow through it become part of what the AI trusts and acts on. A poorly scoped server can give an agent far more access than you intended.