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

# download_attachment

> Mint a short-lived download URL for one email attachment — the bytes go straight to disk, never through the conversation.

Returns a short-lived HTTPS download URL for a single attachment, identified by the message id plus the `blobId` listed by [`search_emails`](/reference/mcp/search-emails), [`get_thread`](/reference/mcp/get-thread), or [`get_email`](/reference/mcp/get-email). The tool result is \~100 tokens of metadata and a URL — **never the file content**; the agent saves the file out-of-band, e.g. with `curl -o`.

## Parameters

<ParamField body="emailId" type="string" required>
  The id of the message the attachment belongs to, as returned by `search_emails` or `get_thread`.
</ParamField>

<ParamField body="blobId" type="string" required>
  The attachment's `blobId`, as listed by `search_emails`, `get_thread`, or `get_email` for that message.
</ParamField>

## Example

```json Call theme={null}
{ "emailId": "Mabc123", "blobId": "Gd41d8cd" }
```

```text Result theme={null}
invoice-2026-06.pdf (application/pdf, 182.4 KB)
Download URL (expires in 24 hours):
https://mcp.langmail.me/attachments/3K9c…Qw

Save it straight to a file, e.g.:
  curl -sS -o 'invoice-2026-06.pdf' 'https://mcp.langmail.me/attachments/3K9c…Qw'
Do not fetch this URL into the conversation — the file may be large.
```

## How the URL works

* The URL is a single-blob capability: an unguessable 256-bit token bound server-side to the granting user's session and exactly one attachment. No bearer token is needed (or accepted) to fetch it — the token *is* the credential, like a presigned URL.
* It expires after **24 hours** and is reusable within that window (retries are fine). After expiry — or a server restart, or once the session's access token lapses — the URL returns a `404`/error; call the tool again for a fresh one.
* `GET` on the URL streams the file with its correct `Content-Type` and a `Content-Disposition` filename; there is no size cap.
* The filename and media type in the result come from the message itself, not from client input.

## Errors

* `Email not found: <emailId>` — no message with that id in the caller's mailbox.
* `No attachment with blobId <blobId> on message <emailId>.` — the blob isn't an attachment of that message.