Agents
Model Context Protocol (MCP)
An open standard that governs how AI applications dock onto tools and data sources: once an MCP server has been built for your own system, all compatible models and agents can use it. That spares the sprawl of bilateral integrations, a kind of USB for AI access, introduced by Anthropic in 2024 and widely adopted.
An example
Before MCP, every combination had to be built individually: model A to the CRM, model A to the ticketing system, model B to the same CRM: four connections for two models and two systems. With a common protocol this becomes one plus one: the CRM exposes its functions once, and any model that speaks the protocol can use them. The same simplification USB brought to peripherals.
A closer look
MCP is an open standard through which a model addresses tools, data, and templates in a uniform way. The benefit is composability: a server built once can be used by different applications instead of being reconnected for each.
Why it matters
The gain in interchangeability is real and the reason the protocol caught on quickly. The security issue comes along with it, though: an MCP server is a door into a business system, and whoever opens it should know who walks through. In practice, individual teams set up such servers on their own, because it takes ten minutes, and nobody in IT security hears of it. Before MCP is spread through the company, two things need to be settled: a directory of which servers exist, and the rule that write access requires approval.
Commonly misunderstood
A protocol settles the form of the connection, not the rights. Whoever connects a server still decides for themselves which data to release and who may steer the agent.