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

# Crawl time and inference time

> AI apps learn about your video at two moments: when a crawler reads your page, and when an agent answers a question. Each moment reads a different format.

An AI app meets your video twice.

* **Crawl time.** A search engine or AI crawler reads your page ahead of any question, and stores what it finds in its own index. It reads structured data: JSON-LD.
* **Inference time.** A person asks a question. The AI app looks for an answer right then. It can read your ARV record, call MCP tools, or use tools your page offers. It reads rich data: every moment, the transcript, and the rights.

JSON-LD is the structured data format search engines read from a `<script type="application/ld+json">` tag in your page. The page block gives you this tag ready made.

```mermaid theme={null}
flowchart TB
  subgraph crawl["Crawl time: before any question"]
    page["Your video page<br/>JSON-LD with chapters"] --> crawler["Search engines and<br/>AI crawlers"]
    crawler --> index["Their index"]
  end
  subgraph infer["Inference time: when someone asks"]
    ask["Person asks an AI app"] --> agent["AI agent"]
    agent -->|"visits your page"| arv["ARV record<br/>via the link tag"]
    agent -->|"connector turned on"| mcp["MCP tools<br/>search, answer, play"]
    agent -->|"agent in a browser"| webmcp["WebMCP page tools"]
  end
  index --> agent
```

## What each one reads

| | Crawl time | Inference time |
| - | - | - |
| Who reads | Googlebot, Bingbot, GPTBot, ClaudeBot and other crawlers | The AI app answering a question |
| Format | schema.org `VideoObject` and `Clip` as JSON-LD, plus your video sitemap | The ARV record (JSON), MCP tools, WebMCP page tools |
| Holds | Title, description, dates, duration, thumbnail, chapters | Every moment, timed transcript, on-screen text, people, places, rights, playback |
| Where it comes from | The [page block](https://docs.agentcdn.com/guides/page-metadata.md) in your page | The links in the page block, the [MCP servers](https://docs.agentcdn.com/guides/connect-mcp.md), and the WebMCP script |
| Needs an install | No | No for the ARV record and WebMCP. Yes for MCP: the person turns on a connector. |
| How fast it shows | Weeks. You do not control when a crawler visits or what it keeps. | At once. An MCP connector works the moment it is added. |

## The three ways in at inference time

* **The ARV record.** The page block adds a `<link rel="agent-ready-video">` tag to your page. An agent that visits the page follows it to the full record. Nothing to install. See [The ARV record](https://docs.agentcdn.com/concepts/arv-record.md).
* **MCP.** MCP (Model Context Protocol) is how AI apps connect to outside tools. A person turns on a connector in ChatGPT, Claude or another app. The agent can then search moments, answer with time-coded citations, and play the moment inside the chat. See [Play a moment in an AI app](https://docs.agentcdn.com/guides/play-in-chat.md).
* **WebMCP.** WebMCP is a browser standard that lets a page offer tools to an AI agent running in the browser. When it is on for your account, the page block adds a small script that offers three tools on your video page: search this catalog, get the context of a moment, and play a moment. Nothing to install.

## What this means for your plan

* Put the page block in your pages first. It helps classic search and AI search at the same time, because both read the same JSON-LD.
* Expect crawl-time results to take weeks to show in AI answers. Measure them with [AI visibility tracking](https://docs.agentcdn.com/guides/visibility.md), before and after.
* Use MCP when you want a result you control today, such as a demo inside ChatGPT or Claude.
* Import the full-length video. AgentTV finds the moments inside it, so you do not need to cut clips for AI to find a single moment.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.