What is MCP (Model Context Protocol)?

Model Context Protocol (MCP) standardizes how an application exchanges tools, resources and prompts with a server. It can connect AI agents to business systems or expose a RAG retriever. It does not choose the model, implement retrieval or authorize a business action by itself.

Portability is conditional. Check each client’s supported protocol revision, transport, authorization flow and features. The 2026-07-28 revision removed the initialization handshake and protocol sessions in favor of per-request metadata. Modern-only and legacy-only peers need an explicit compatibility path; an MCP label alone does not make them interoperable.

Hypothetical example: a company exposes CRM lookup and document retrieval through MCP. A second host may reuse those servers, but the team must verify tool discovery, schemas, authentication, tenant boundaries and approval behavior in that host. The versioning and compatibility specification distinguishes the legacy and modern eras. Reusing server logic can reduce integration work without eliminating client validation.

MCP includes an optional authorization framework, not a complete application permission policy. An HTTP deployment using that framework must validate tokens intended for that server; it must not pass a client token through to an upstream API. Enforce tenant, object and action permissions at the data boundary, and require appropriate approval for consequential actions. Before starting a third-party setup, inspect its command, paths, secrets, pinning and network exposure with the local MCP Install Risk Scanner.