Model Context Protocol (MCP)
As of:
The Model Context Protocol (MCP) is an open standard that defines how AI applications connect to external tools and data sources. Introduced by Anthropic in late 2024 and since adopted across the industry, it solves a practical problem: previously, every integration between a model application and a third-party system had to be built individually.
Technically, MCP follows a client-server architecture. An MCP server wraps a system — a database, a file system or a SaaS service, say — and exposes its capabilities as tools, resources and prompts through a uniform JSON-RPC interface. The host application (a chat client, an IDE or an AI agent) connects as a client and passes the tools on to the LLM. The connection runs locally via standard input/output or remotely via HTTP. A server written once works with every MCP-capable application — hence the frequently used comparison to a “USB-C port for AI”.
In practice, MCP significantly lowers integration effort and has become the de facto standard for connecting tools. Two misconceptions are common: first, MCP does not make a model smarter — it only standardises access; whether tools are used sensibly still depends on the model and the prompt. Second, remote MCP servers in particular are security-relevant: permissions and prompt injection through tool outputs have to be actively addressed.