Show HN: MCP is for tools. A2A is for agents. What's for websites? (www.rtrvr.ai)

A new paper from rtrvr.ai introduces the Agent-Web Protocol Stack, highlighting an urgent need for a dedicated execution layer for AI agents interacting with the web. As AI agents evolve to independently navigate, transact, and act online, current web protocols (HTTP, HTML, etc.) remain tailored for human users, limiting agents' capabilities. The study outlines various agent architectures, including text-based and DOM-based agents, while emphasizing Rover, rtrvr.ai's DOM-native web agent SDK, which aims to facilitate structured interactions by providing clearer integration with web content. The significance of this development lies in its potential to reshape the web's infrastructure for an increasingly AI-driven ecosystem. The paper identifies that existing agents lack essential features for engaging with dynamic web content, such as executing JavaScript or maintaining session states. By establishing protocols for agent-to-agent (A2A) and agent-to-tool (MCP) communication, alongside innovations like the proposed Pay Per Crawl model—enabling identity-based payment for web access—this framework aims to enhance the interaction and functionality of AI in online environments. Ultimately, these advancements could drive a transformative shift in how data is accessed and utilized on the web, paving the way for more efficient AI applications.