The agentic web is evolving with two distinct approaches for artificial intelligence interaction: identity, represented by the proposed llms.txt standard, and capability, embodied by the Web Model Context Protocol (WebMCP).
This divergence highlights a fundamental debate over how AI agents should interact with websites, with one method gaining widespread, often unintended, adoption while the other requires deliberate implementation but offers direct utility for task completion.
Jeremy Howard, co-founder of Answer.AI, proposed llms.txt as a plain-text file designed to offer AI models a human-curated index of website content.
However, John Mueller of Google expressed skepticism regarding the effectiveness of llms.txt for AI systems, according to reports from Search Engine Journal.
Conversely, WebMCP is a browser standard developed by Google and Microsoft through the W3C Web Machine Learning Community Group.
The protocol allows websites to expose callable tools, enabling AI agents to directly perform tasks and complete transactions.
AIOSEO, a WordPress plugin, currently generates llms.txt by default for more than 3 million websites globally.
This widespread adoption has occurred largely without deliberate choice from website owners or confirmed usage by major AI systems, according to industry observations.
Google’s WebMCP is undergoing a public origin trial in Chrome, from version 149 through 156, with Google’s Gemini AI system already consuming these exposed tools within the Chrome browser.
Proponents of WebMCP argue that it offers a stronger foundation for AI task completion and transactions by focusing on direct capability, rather than the identity-focused, brochure-like approach of llms.txt.
The W3C Web Machine Learning Community Group continues to oversee the development of WebMCP, aiming to standardize how AI agents can access and utilize website functionalities.
Source: Search Engine Journal
