MCP needs a way for servers to push prompt context to clients.
Today, if I select something in Figma, I have to perform a second action and tell the agent: "hey, I selected something in Figma". Could also be used for CI failures, exception pages, text highlighting, etc.
Yes, there are security concerns but these could use the same permission model as tool calls... but it may also be safer? With push, users can review the exact content being added to the prompt, while for tool calls you review only the inputs, not the output.