Agents
Model Context Protocol (MCP)
MCP is an open protocol that lets an AI client connect to servers exposing tools, resources and prompts through a standard interface.
Also written: MCP
The value is that a capability implemented once as an MCP server works with any client that speaks the protocol, instead of being reimplemented per framework.
Configuration is a JSON block naming each server, the command that starts it and its arguments. Arguments must be an array rather than a string, because the command is spawned directly rather than through a shell.
In practice
The value is not the protocol, it is that a tool integration written once works with any client that speaks it. The security position is unchanged though: an MCP server is code with credentials, and anything it returns to the model is untrusted input that can carry instructions. Treat a tool result exactly as you would treat a scraped web page.
Common questions
What problem does MCP solve?
Writing a tool integration once instead of once per client. Before it, connecting a data source to three different assistants meant three implementations of the same thing.
Is MCP secure by default?
No more than any other code with credentials. An MCP server runs with whatever access it was given, and anything it returns is untrusted input that can carry instructions. Treat a tool result exactly like a scraped web page.