Model Context Protocol

MCP uses a client-server design. An MCP server wraps a system, such as a file store, database, ticketing tool or web API, and describes what it offers: tools the model can call, resources it can read and reusable prompts. An AI application acts as the host, running an MCP client that connects to one or more servers, lists their capabilities and passes them to the model, which can then request actions through tool calling. Messages use JSON-RPC.

Why it matters: before MCP, each AI application needed its own custom integration for every tool. A shared protocol means an integration built once can work across many assistants, IDEs and agent frameworks, and buyers can check whether a product supports MCP to judge how easily it connects to their systems.

Example: a developer installs an MCP server for their company's issue tracker; their AI coding assistant can then read tickets and update their status during a coding session.

Limits: MCP standardises the connection, not the safety of what is connected. Servers can expose sensitive data or powerful actions, so install only trusted servers, limit their permissions and watch for prompt injection through content the servers return.