Do attachments come back?

The short answer

The text of a file inside another file comes back. This holds for a ZIP, an email attachment and a file attached to a PDF.

A ZIP also returns the path of each file, as a line before its text. An Outlook .msg returns the attachment name too.

An .eml does not return its headers. From, To, Cc, Subject and Date are all missing. An .mbox does not return them either.

Email and PDF attachments come back without a file name.

metadata.ocr stays false for a picture inside a ZIP, although the picture is read.

The engine is tika-server-standard 3.3.1.

Source: terraform/lambda-layers/build/versions.env (versions.env: TIKA_VERSION), accessed 2026-10-19.

What comes back, case by case

We sent each file to the production API. Every inner file holds one marker sentence, so you can see which text came from which file. Every row shows its source. The inputs and the raw responses are in the repo.

What this means for your pipeline

To get one text per file from a ZIP, ask for format=json. Split the elements on the level-1 headings. Each heading is a file path. A nested ZIP gives a flat list, so a path does not tell you the parent.

If you need the file name of an email attachment, send the attachments one by one. You hold the name, so you can keep it beside the text.

If you need the From, To or Subject of an .eml, read them yourself. The next section shows how. An Outlook .msg already returns them.

Do not read metadata.ocr as proof that OCR did not run. Check the text instead. Text in pictures and scans says where OCR runs. Quality checks lists the checks to run on the output.

How to get the missing parts yourself

Run each snippet next to your txtfetch call. Each snippet ran on the sample file named under it.

Read the headers and the attachment names of an .eml

The API drops both. The Python standard library reads them from the file you hold.

Python, run on mail.eml

import email
from email import policy

with open("mail.eml", "rb") as f:
    msg = email.message_from_binary_file(f, policy=policy.default)

for name in ("From", "To", "Cc", "Subject", "Date"):
    print(name + ":", msg[name])

for part in msg.iter_attachments():
    print("attachment:", part.get_filename(), part.get_content_type())

It printed

From: Ana Silva <ana@example.com>
To: Ben Okoye <ben@example.com>
Cc: Chi Wang <chi@example.com>
Subject: Quarterly files
Date: Mon, 19 Oct 2026 09:00:00 +0000
attachment: mail-report.pdf application/pdf
attachment: mail-memo.docx application/vnd.openxmlformats-officedocument.wordprocessingml.document
attachment: mail-scan.png image/png
attachment: mail-bundle.zip application/zip

Split a ZIP response into one text per file

Ask for format=json. Each level-1 heading is a file path. Group the elements after it. This snippet reads the saved ZIP response.

Python, run on responses.json

import json

with open("responses.json", encoding="utf-8") as f:
    response = json.loads(json.load(f)["archive-zip"]["json"])

files = {}  # path -> text
current = None
for element in response["elements"]:
    if element["type"] == "heading" and element.get("level") == 1:
        current = element["text"]
        files[current] = []
    elif current is not None and element["text"]:
        files[current].append(element["text"])

for path, parts in files.items():
    print(path, "->", " ".join(parts) or "(no text)")

It printed

notes.txt -> Text from notes.txt.
pages/page.html -> Text from page.html.
scan.png -> Picture text in scan.png
memo.docx -> Text from memo.docx.
report.pdf -> Text from report.pdf.
inner.zip -> (no text)
deep.txt -> Text in the inner zip.

What we did not test

We tested uploaded files. We did not test a page that txtfetch fetches from a URL. We did not test .pst or .ost mailboxes, or formats such as 7z, tar and RAR. We make no claim about them. We tested a ZIP with an encrypted member. It returns success with the member name only. See Password-protected files.

Read the ZIP page, the email page and the Outlook .msg page for the rest of each format.

Frequently asked questions

Does txtfetch read the files inside a ZIP?

Yes. It returns every file, in archive order. A ZIP inside a ZIP is walked too. A picture inside a ZIP is read by OCR.

Can I tell which file a text came from?

For a ZIP, yes. The path of each file comes back as its own line before its text. With format=markdown or format=json it is a level-1 heading. For email and PDF attachments, no. The file name does not come back.

Do the From, To and Subject lines of an .eml come back?

No. In our probe of production only the body and the attachment text came back. Read the headers from the file with your own code. An Outlook .msg is different. It returns the subject, the sender name, To and Cc.

Does an .mbox come back with its headers?

No. Only the message bodies come back. There is no header and no line between two messages.

Does metadata.ocr tell me a picture inside a ZIP was read?

No. In our probe the picture inside a ZIP was read by OCR, but metadata.ocr said false. Do not use that flag to decide whether OCR ran on an archive.

Why does an .eml with a ZIP attached give me one paragraph?

In format=markdown and format=json the entry name and its text join in one paragraph. Use the plain text output, or send the ZIP on its own, if you need the file names apart.

Does a PDF with an attached file come back whole?

The text of the attached file comes back after the page text. Its file name does not come back.