What is the Model Context Protocol (MCP)?

Model Context Protocol is an open standard introduced by Anthropic in 2024. It defines a common way for AI applications, such as chat assistants and coding tools, to connect to other systems. Instead of every AI app building a custom connection to every tool, a tool can offer one connector, called an MCP server, and any AI application that supports MCP can use it.

For example, a company could connect its AI assistant to its project tracker, shared drive, and CRM through MCP servers. An employee could then ask the assistant which deals are waiting on a proposal, and it would check the CRM, find the relevant documents, and answer, or, with permission, update a task, all from within the chat.

MCP is often compared to a universal adapter: it standardizes the plug, not what is on the other end. Each connection can offer data to read, actions to take, or both. The protocol itself does not make a connection safe or accurate; that depends on the specific server, how it was built, and what access it has been given.

Because MCP servers can let an AI read private data and take real actions, they deserve the same caution as any integration. Connections should come from trusted sources, use the narrowest access that works, and require approval for actions that change or send things. Content fetched through a connection, such as a web page or an email, can also contain hidden instructions aimed at the AI, a risk known as prompt injection.

An example

A team connects its assistant to the company's ticketing system through an MCP server, so it can summarize open issues on request.