MCP is an open standard, introduced by Anthropic in November 2024, that defines how AI applications connect to external data and tools - one common plug instead of a custom integration for every tool and every AI. Vendors talk about it because it turns "we integrate with your AI" from a months-long project into a checklist item.
The problem it solves
Every useful AI feature needs context: your documents, tickets, calendar, customer records. Before MCP, connecting each data source to each AI product meant a bespoke integration on both sides - the so-called M-times-N problem. MCP standardises the interface: a data source or tool exposes one MCP server, and any MCP-compatible AI can use it.
What it looks like in practice
An MCP server wraps a system - say your file storage, your CRM, or an internal database - and describes what the AI may read or do there. The AI application (the MCP client) connects, discovers the available tools, and calls them with a defined format. The current specification is public and versioned, and the ecosystem has grown well beyond Anthropic: OpenAI, Google DeepMind and Microsoft adopted MCP in 2025, making it the closest thing the industry has to a default connector standard.
What it is not
MCP is a connector, not a guardrail. It standardises how an AI reaches your data - not what it may do once connected. Permissions, access control and approval steps still have to be designed and enforced in your systems and in the agent's harness. An MCP server that exposes too much is an efficiency gain for an attacker too.
Why a Swiss SME should care
If your tools speak MCP, you stop betting on one AI vendor: the assistant you choose this year can be swapped next year without rebuilding every integration. When you evaluate AI products, "does it support MCP" is now a fair checklist question - and "which MCP servers do you already have" tells you how fast a pilot can start.
Sources
Last updated: 20 October 2026.
