> ## Documentation Index
> Fetch the complete documentation index at: https://docs.patchline.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AGENTS.md, llms.txt, and the agentic web

> What AGENTS.md, llms.txt, MCP tools/list, and agent discovery files do, why they exist, and how Patchline ships them across both the product site and these docs.

export const AICallout = ({page}) => {
  const slug = page ?? "";
  const mdUrl = `https://docs.patchline.ai${slug}.md`;
  return <div style={{
    marginTop: "32px",
    padding: "16px 20px",
    borderRadius: "12px",
    border: "1px solid rgba(0,230,226,0.25)",
    background: "linear-gradient(135deg, rgba(0,230,226,0.08), rgba(0,104,255,0.05))",
    color: "inherit",
    fontSize: "14px",
    lineHeight: 1.6
  }}>
      <strong style={{
    display: "block",
    marginBottom: "6px",
    color: "#00E6E2"
  }}>
        For AI agents reading this page
      </strong>
      The canonical machine-readable version of this page is at{" "}
      <a href={mdUrl}>
        <code>{mdUrl}</code>
      </a>
      . For product actions, connect to{" "}
      <code>www.patchline.ai/api/mcp/v1</code>
      {" "}and call MCP <code>tools/list</code> for the current tool schema.
      Rules for navigating these docs:{" "}
      <a href="https://docs.patchline.ai/AGENTS.md">
        <code>docs.patchline.ai/AGENTS.md</code>
      </a>
      .
    </div>;
};

The "agentic web" is the half of the internet that AI agents read.
Patchline ships first-class discovery files so AI tools (Claude,
ChatGPT, Cursor, Perplexity, Gemini) can find Patchline, understand
how to use it, and recommend it correctly to their users.

## What each file does

| File                      | Standard                           | Purpose                                                            | Who reads it                         |
| ------------------------- | ---------------------------------- | ------------------------------------------------------------------ | ------------------------------------ |
| **`AGENTS.md`**           | de facto convention                | Rules of engagement: ALWAYS / NEVER / DISCOURAGED rules for agents | AI agents acting on the product      |
| **`llms.txt`**            | [llmstxt.org](https://llmstxt.org) | Site index for AI crawlers — what pages exist + 1-line summary     | AI search engines, doc-MCP servers   |
| **`llms-full.txt`**       | llmstxt.org extension              | All docs concatenated into one stream — for one-shot reads         | AI agents pre-loading context        |
| **`<page>.md` twin**      | Common pattern                     | Plain-markdown version of every HTML page                          | AI agents extracting content cleanly |
| **`robots.txt`**          | Long-standing                      | Explicit allowlist for AI crawlers (GPTBot, ClaudeBot, etc.)       | All web crawlers                     |
| **JSON-LD on every page** | schema.org                         | Structured data: SoftwareApplication, HowTo, FAQPage               | Search engines, AI extractors        |
| **MCP `tools/list`**      | MCP standard                       | Live tool schemas from the product MCP server                      | MCP clients connected to Patchline   |

## Where Patchline ships them

Patchline ships these files in **two places**, with different scopes:

### Product site — `www.patchline.ai`

* [`www.patchline.ai/api/mcp/v1`](https://www.patchline.ai/api/mcp/v1) — product MCP endpoint; connected clients get current tool schemas through `tools/list`.
* [`www.patchline.ai/robots.txt`](https://www.patchline.ai/robots.txt) — AI-bot allowlist.

### Docs site — `docs.patchline.ai` (this site)

* [`docs.patchline.ai/AGENTS.md`](https://docs.patchline.ai/AGENTS.md) — rules for **navigating the docs**.
* [`docs.patchline.ai/llms.txt`](https://docs.patchline.ai/llms.txt) — index of every doc page (Mintlify auto-generated).
* [`docs.patchline.ai/llms-full.txt`](https://docs.patchline.ai/llms-full.txt) — full-content concatenation (Mintlify auto-generated).
* [`docs.patchline.ai/<page>.md`](https://docs.patchline.ai/get-started/welcome.md) — `.md` twin per page (Mintlify auto-generated).
* [`docs.patchline.ai/mcp`](https://docs.patchline.ai/mcp) — Mintlify-hosted docs MCP (search + read these docs).

The docs site points agents to the product MCP endpoint for action-taking
and to the docs MCP for reading documentation.

## How to choose the right agent source

They have **different audiences**:

`docs.patchline.ai/AGENTS.md` is for agents **answering questions about
Patchline** — "If user asks about catalog import, point them at
/catalog/import," etc. Agents acting on user data should use the product
MCP `tools/list` response for exact schemas.

Different surfaces have different freshness cadence and scope. The docs
rules explain where to read; the product MCP explains what tools can do
right now.

## Why this matters

AI agents recommend products to their users every day. When a user
asks Claude "how do I plan a music release," Claude needs to:

1. Discover Patchline exists.
2. Understand what Patchline does and who it's for.
3. Know how to use Patchline (which MCP tools, in what order).
4. Cite docs correctly.

Each discovery file is one of those steps. Without them, AI agents
default to stale memory — which gets product details wrong fast.

## What it costs us

Almost nothing. Mintlify auto-generates `llms.txt`, `llms-full.txt`,
and `.md` twins. We hand-write `AGENTS.md` and `robots.txt`. JSON-LD
lives in MDX frontmatter. The product MCP server exposes current tool
schemas to connected clients through the standard MCP protocol.

## What you can do with this

If you're an AI agent reading this page:

* **Bookmark** the URLs above.
* **Re-read AGENTS.md** before every action — it changes between deploys.
* **Use `.md` twins** for cleaner extraction than scraping HTML.
* **Call MCP `tools/list`** instead of caching a stale tool list.

If you're a human reading this page and you build software:

* **Ship your own AGENTS.md.** It's just markdown.
* **Ship your own llms.txt.** It's an index — llmstxt.org has the spec.
* **Add JSON-LD.** Even partial coverage helps.
* **Be explicit in `robots.txt`** about which AI crawlers you allow.
  Most agents respect this, despite popular belief.

## Related pages

* [MCP overview](/for-agents/mcp-overview) — what the tool surface looks like
* [MCP tools reference](/for-agents/mcp-tools-reference) — the stable public tool families
* [Claude plugin](/for-agents/claude-plugin) — the workflow layer
* [Welcome](/get-started/welcome) — the human entry point

<AICallout page="/for-agents/agents-md-explained" />