llms.txt vs MCP: what's the difference and which do you actually need?

Two specs are fighting for the "how should AI agents understand your website" mindshare right now: llms.txt and the Model Context Protocol (MCP). If you've been in dev Twitter lately you've seen both mentioned in the same breath, usually by people who seem to assume everyone already knows the difference. They don't. Here's the actual breakdown.

What llms.txt is

llms.txt is a file you put at the root of your website — like robots.txt, but for language models. It's a plain-text summary of what your site is, what it does, and links to the most important pages. The idea is that when an LLM is trying to understand your product, it can pull this file instead of having to scrape and parse your entire site.

A basic llms.txt looks like this:

# YourProduct

> One-sentence description of what your product does.

## Docs
- [Getting started](https://yoursite.com/docs/getting-started)
- [API reference](https://yoursite.com/docs/api)
- [Authentication](https://yoursite.com/docs/auth)

## About
- [Pricing](https://yoursite.com/pricing)
- [Changelog](https://yoursite.com/changelog)

It's passive. You publish it, and LLMs that know to look for it will find it. You don't control when or how it's read.

What MCP is

MCP (Model Context Protocol) is a protocol for tools. Specifically, it lets AI agents call functions — and one of those functions can be "query this website". Unlike llms.txt, MCP is interactive. The agent asks a question, the server looks up the answer in indexed content, and returns a cited response.

An MCP server exposes tools. AgentReady's MCP server exposes three: submit_site, list_sites, and ask_site. When Claude wants to know something about your docs, it calls ask_site("yourdomain.com", "how do I authenticate?") and gets an answer back.

The actual difference

llms.txt is a hint. MCP is a query. llms.txt says "here's what we are and here are some links". MCP says "ask me anything and I'll look it up". One is a static file, the other is a live API.

They're also used at different points. llms.txt gets read during training or when an LLM is building context about your product in general. MCP gets called at runtime when a user is actively working and needs a specific answer.

Which one do you need?

Honestly, both. They serve different purposes and the effort for each is low.

Do llms.txt if: you want your product to be understood correctly by LLMs during training and general context-building. It's a one-time 20-minute job. Put it at /llms.txt and you're done.

Do MCP if: you want users to be able to query your docs directly from Claude Desktop, Cursor, or any other MCP client. This is where the interactive value is — someone asks a question, they get your answer, with a citation. That's a materially better experience than "I don't know, check the docs".

If you want both without building either from scratch, AgentReady generates an llms.txt from your site when it indexes it, and serves your content via MCP automatically.

The short version

llms.txt = a static hint file for LLMs. MCP = a live query interface for AI agents. You want both, they take different amounts of effort, and they're not competing standards — they solve different problems at different points in the AI interaction lifecycle.