An allow-list is a security control in which the entities permitted to access or run (users, applications, IP addresses, domains, commands, and so on) are defined in advance, and anything not on the list is denied. Because the default is to deny, it can block unknown threats as well, but the list must be updated whenever a new entity is needed. It is the opposite of a deny-list (blocklist), which enumerates only what is forbidden, and was formerly called a whitelist.
In AI agent environments, allow-lists are used to restrict which MCP servers an agent may connect to, which tools it may call, and which models and providers it may use.