agent.json: A New Open Protocol for AI Agents to Communicate with Websites

Summary

A developer has introduced agent.json, an open protocol designed to enable standardized communication between AI agents and websites, similar to how robots.txt governs web scraping. The protocol addresses a critical gap in AI-to-website interaction by providing three key endpoints: a discovery endpoint that reveals available actions and authentication requirements, a message inbox for agents to submit requests, and a polling mechanism for retrieving responses.

The protocol uses JSON Schema to define action parameters, ensuring agents understand exactly what data to submit for tasks like booking appointments or requesting refunds. It supports multiple response patterns including synchronous replies, polling, and webhook callbacks with HMAC-signed authentication. The implementation is language-agnostic and can be deployed on any backend framework, with a reference implementation provided via Cloudflare Workers that includes AI-powered request classification and email routing capabilities.

Editorial Opinion

agent.json addresses a genuinely important infrastructure problem as AI agents become more capable and prevalent. The analogy to robots.txt is apt—this could become a widely-adopted standard if it gains traction. However, adoption will depend on whether websites see sufficient value in making themselves agent-friendly and whether the protocol proves robust enough for real-world use cases with authentication, rate limiting, and error handling at scale.