Abstract

In the era of AI agents, seamless interaction between large language models (LLMs) and web app APIs remains a critical challenge due to inconsistent documentation and resource intensive protocols like Anthropic’s Model Context Protocol (MCP). We propose LLM-API.txt, a lightweight, text-based protocol inspired by robots.txt, designed to enable web apps to provide structured, machine-readable API specifications. Hosted at a standard URL, LLM-API.txt outlines endpoints, parameters, authentication, and rate limits, empowering LLMs to execute complex API calls with minimal developer overhead. Unlike MCP, which burdens web apps with complex server requirements, our protocol requires only a static file, democratizing AI agent access and reducing costs. We evaluate LLM-API.txt using small language models and state-of-the-art LLMs, testing authenticated API calls, rate limit compliance, and edge cases like malformed files. Our results demonstrate robust, efficient API interactions, paving the way for a standardized, ML-free web where LLMs handle the heavy lifting, with potential to redefine AI-driven web app interactions.