Show HN: Agent Action Protocol (AAP) – MCP got us started, but is insufficient (github.com)

The Agent Action Protocol (AAP) has been introduced as a draft specification to enhance the capabilities of AI agents, which are increasingly taking independent actions like creating tickets, sending emails, and deploying code. While the existing Model Communication Protocol (MCP) made strides in standardizing how these agents discover and utilize tools, it fell short in defining crucial aspects of action management, such as action identity, lifecycle hooks, permissions, and error recovery. AAP aims to fill these gaps by providing a comprehensive framework that standardizes how actions are defined, executed, and monitored, making it an essential step for the evolving AI/ML landscape. Significantly, AAP introduces several key concepts, such as a declarative YAML manifest for action capabilities, typed input/output schemas, and support for state management and error recovery—features that facilitate robust action lifecycle management. This standardized approach not only promotes composability and resource control but also enhances observability by ensuring that every action execution is traceable. Touted as "the package.json for agent actions," AAP sets the stage for a more structured and secure ecosystem for AI agents. The draft invites input from the community, emphasizing collaboration to refine its specifications and address the pressing needs of developers and infrastructure teams working with AI agents.