---
title: document_get
description: Fetch one brand document — images come back as an image the assistant can actually look at.
---

Fetches one **document**. The difference from [`document_list`](/mcp/tools/document-list) is what you get back for an image: the picture itself, not just a link to it.

That's the point of this tool. An assistant that can _see_ your logo can tell you whether a headline sits well beside it; one holding a URL can only take your word for it.

## Input

| Field         | Type     | Required | Description                      |
| ------------- | -------- | -------- | -------------------------------- |
| `document_id` | `string` | yes      | The id of the document to fetch. |

## Output

Always the metadata — `{ id, name, filename, mimeType, sizeBytes, url }`.

**Images** under 1 MB additionally come back as an image the assistant can view. Everything else — PDFs, larger images — returns metadata and the URL alone.

:::info[Why PDFs don't come back as readable text]

Reading the _contents_ of a PDF needs a parser, which is its own piece of work and isn't built yet. Handing an assistant the raw file wouldn't help it — so a PDF comes back as a link, and an assistant with its own fetching ability can still go and read it.

:::

## Example

> _"Does this headline work with our logo?"_

```ts
document_get({ document_id: 'doc_…' });
// → metadata, plus the logo itself for the assistant to look at
```

Requires the `brand:read` scope.