# AI media pipeline

Wire agents into your media library: search by meaning, pull references, generate variations, and file the results back — automatically. Two paths: call the API directly, or point any MCP client at the hosted server.

![An AI media pipeline using the Playbook MCP server](/img/examples/ai-pipeline.webp)

Live preview

**[Open the interactive preview →](/demos/ai-pipeline/index.html)**  ·  a self-contained `index.html` you can click through in your browser (open photos, favorite shots, switch tabs) or download.

## The flow[​](#the-flow "Direct link to The flow")

1. Agent **searches semantically** for references (AI search).
2. Agent **ingests generated variations** in bulk.
3. Agent **auto-tags** them with custom fields.
4. Or skip the glue code entirely and use the **MCP server**.

## 1. Semantic search for references[​](#1-semantic-search-for-references "Direct link to 1. Semantic search for references")

AI search returns meaning-ranked results with a cursor for paging deeper:

agent.ts

```
const ORG = "your-org";

const TOKEN = process.env.PLAYBOOK_TOKEN!;



type SearchPage = { parentId: string; afterCursor: string };



async function aiSearch(query: string, page?: SearchPage) {

  const params = new URLSearchParams({ query });

  if (page) {

    params.set("parent_id", page.parentId);

    params.set("after_cursor", page.afterCursor);

  }

  const res = await fetch(

    `https://api.playbook.com/v1/${ORG}/ai_search?${params}`,

    { headers: { Authorization: `Bearer ${TOKEN}` } },

  );

  return res.json(); // { data: [...assets], pagy: { parent_id, after_cursor, has_next_page } }

}



const refs = await aiSearch("brand palette: warm coral, moody lighting");



if (refs.pagy.has_next_page) {

  const moreRefs = await aiSearch("brand palette: warm coral, moody lighting", {

    parentId: refs.pagy.parent_id,

    afterCursor: refs.pagy.after_cursor,

  });

}
```

## 2. Ingest generated variations[​](#2-ingest-generated-variations "Direct link to 2. Ingest generated variations")

Your model produces new images (hosted somewhere reachable). Add them in one call — Playbook fetches and processes each in the background:

ingest.ts

```
async function ingestVariations(boardToken: string, urls: string[]) {

  const res = await fetch(

    `https://api.playbook.com/v1/${ORG}/assets/batch_create_from_urls`,

    {

      method: "POST",

      headers: {

        Authorization: `Bearer ${TOKEN}`,

        "Content-Type": "application/json",

      },

      body: JSON.stringify({

        batch: {

          collection_token: boardToken,

          assets: urls.map((uri, i) => ({

            uuid: `gen-${i}`,

            uri,

            title: `Variation ${i + 1}`,

          })),

        },

      }),

    },

  );

  return (await res.json()).data; // [{ uuid, asset }] — each starts as a skeleton

}
```

## 3. Auto-tag the results[​](#3-auto-tag-the-results "Direct link to 3. Auto-tag the results")

Have the agent file each variation against your taxonomy:

tag.ts

```
async function tag(assetToken: string, fields: Record<string, string>) {

  await fetch(`https://api.playbook.com/v1/${ORG}/assets/${assetToken}/update`, {

    method: "PATCH",

    headers: {

      Authorization: `Bearer ${TOKEN}`,

      "Content-Type": "application/json",

    },

    body: JSON.stringify({ asset: { fields } }),

  });

}



await tag("variation-1", { "Approval Status": "In Review", Source: "AI" });
```

## 4. Or skip the code — use MCP[​](#4-or-skip-the-code--use-mcp "Direct link to 4. Or skip the code — use MCP")

If your agent runs in Claude, Codex, Cursor, or ChatGPT, you don't need any of the above. Point it at the hosted MCP server and it can search, tag, upload, and organize your media in natural language — sign in with Playbook in the browser, no token needed:

```
claude mcp add --transport http playbook https://mcp.playbook.com/mcp
```

See the [MCP guide](/docs/guides/mcp.md) for the Claude Code plugin, Codex, Cursor, and ChatGPT setup and the full tool list.

## Related[​](#related "Direct link to Related")

* [AI Search](/docs/guides/ai_search.md)
* [Uploading Assets](/docs/guides/upload.md)
* [Custom Fields](/docs/guides/custom_fields.md)
* [MCP Server](/docs/guides/mcp.md)