Fix pypdf "EOF marker not found" (PdfReadError)

The error

Depending on your version and settings, you’ll see one of these:

pypdf.errors.PdfReadError: EOF marker not found

WARNING: EOF marker not found
pypdf.errors.PdfStreamError: Stream has ended unexpectedly

A valid PDF must end with the line %%EOF. pypdf reads the file from the end to find it; if it isn’t there, the file is incomplete or isn’t a PDF at all.

We reproduced both messages with pypdf 6.19: the default mode logs the warning and then raises PdfStreamError, while PdfReader(..., strict=True) raises PdfReadError: EOF marker not found. Older PyPDF2 raised PdfReadError by default.

The three usual causes

1. Truncated download. The connection dropped, a size limit cut the file off, or you read the response before it finished.
2. It isn’t a PDF. A login page, “access denied” page or 404 page was saved with a .pdf name. pypdf then also logs invalid pdf header: b'<!DOC'.
3. The file is still being written by another process, such as a scanner, a sync client or a job that hasn’t finished.

Step 1: check what you actually have

import os

def looks_like_pdf(path):
    with open(path, "rb") as f:
        return f.read(5) == b"%PDF-"

print(os.path.getsize("input.pdf"), looks_like_pdf("input.pdf"))

• 0 bytes or a few KB for a document that should be larger means a failed download. (An empty file raises EmptyFileError: Cannot read an empty file instead.)
• Doesn’t start with %PDF- means it’s something else. Open it in a text editor; you’ll often see HTML.

Step 2: download PDFs properly

Most cases come from download code that doesn’t check the response:

import requests

resp = requests.get(url, timeout=60)
resp.raise_for_status()                      # fail on 403/404/500 instead of saving the error page
if not resp.content.startswith(b"%PDF-"):
    raise ValueError(f"Not a PDF: {resp.headers.get('Content-Type')}")

with open("input.pdf", "wb") as f:
    f.write(resp.content)

If the server needs cookies or a login, you’ll get an HTML page with status 200. The %PDF- check catches that too.

Step 3: repair a damaged file

If the file is a real but truncated PDF, PyMuPDF can often rebuild it:

import pymupdf

doc = pymupdf.open("truncated.pdf")
print(doc.page_count, doc.is_repaired)   # 3 True
doc.save("repaired.pdf", garbage=3, deflate=True)

In our test, a PDF cut off at 50% opened in PyMuPDF with is_repaired == True, all pages and text intact, and the saved copy then loaded in pypdf even with strict=True.

qpdf (qpdf damaged.pdf repaired.pdf) is another option for files with a broken cross-reference table, but it’s less forgiving with truncation. On the same half-truncated test file, qpdf 11.9 gave up with unable to find trailer dictionary while recovering damaged file, while PyMuPDF recovered everything. Try PyMuPDF first.

If neither tool can open the file, the missing data is simply gone. Get a fresh copy from the source.

Related

• PyMuPDF tutorial
• Not a developer? See PDF won’t open? How to fix it.