AgentGrade

AgentGrade is an online tool designed to assess how prepared a website is for interaction with autonomous AI agents. It addresses the need for site owners to understand which protocols, discovery files, and agent-facing features their sites expose, providing a perspective on what AI agents—such as LLM-driven clients—can see and utilize when visiting a site. By scanning a given URL, AgentGrade generates a graded report that highlights the presence or absence of key agent-relevant capabilities, helping site owners identify and address potential gaps.

The platform checks for a variety of discovery files, including llms.txt, llms-full.txt, robots.txt, sitemap.xml, humans.txt, and agent skill manifests. It also verifies Model Context Protocol (MCP) endpoints, transport methods, and tool surfaces, as well as payment protocols such as x402, L402, and Stripe Tempo/MPP. Additional analysis covers published OpenAPI specifications, server URLs, probing of paid endpoints, and the presence of GraphQL surfaces. AgentGrade examines content negotiation capabilities, such as support for JSON and Markdown variants, handling of agent-friendly Accept headers, and the risks posed by JavaScript rendering. It further reviews identity and standards signals, including A2A, WebMCP, structured data, canonical metadata, and security.txt files.

To use AgentGrade, users simply enter a website URL to initiate a free, instant scan. The tool then probes the site from the perspective of an AI agent, fetching, parsing, and verifying each declared capability. Upon completion, it provides a readiness score along with a detailed remediation guide for each check, including links to relevant knowledge-base articles to help site owners address identified issues.

AgentGrade is delivered as a web-based service and does not require installation. The scan and grading process is free, enabling site owners to quickly gauge and improve their website's compatibility with autonomous AI agents.