# <script type="text/llms.txt"> By Steven Van · 2025-08-20 Vercel's default 401 page now embeds a hidden script tag telling AI agents how to fetch a bypass token via its MCP server. Vercel is proposing a convention for putting instructions to AI agents directly inside a page's HTML: a <script type="text/llms.txt"> tag. Browsers ignore script elements with an unrecognised type, so the content has no effect on normal rendering, but an LLM reading the response will still see it. The idea came from a specific problem at [Vercel](): preview deployments are protected by Vercel Authentication by default, which also blocks coding agents like Cursor, Devin, or Claude Code from reaching them. Vercel already runs an MCP server with functions such as get\_access\_to\_vercel\_url and web\_fetch\_vercel\_url to get around that, but agents had no way to know those functions existed. Embedding instructions in the HTML of the HTTP 401 response gives them a way to find out. Vercel has [shipped this]() on its default 401 page for deployments, with a script block telling agents to fetch a Vercel authentication bypass token through the MCP server at mcp.vercel.com, or to prompt the user for one using Vercel's documented bypass method. The convention builds on the existing llms.txt format rather than AGENTS.md, since Vercel treats AGENTS.md as scoped to in-repository agent context and llms.txt as scoped to web publishing. Vercel says it isn't proposing a formal standard: the approach worked as soon as it shipped, without needing support from any LLM provider, and it suggests the same tag could point agents to MCP servers or route error pages toward observability tools. [![]()]() ![Vercel]() Vercel The platform for frontend developers — deploy, preview, and scale web apps and AI agents with zero config. [View Vercel →]() [Original source]() [Read <script type="text/llms.txt"> on Creators Toolbox]() --- Canonical source: https://creatorstoolbox.com/blog/vercel-script-type-text-llms-txt