Repair PDF
'This file can't be opened', 'the file is damaged', a PDF that half-loads and dies: usually the internal structure broke, and usually it can be rebuilt. Upload the file, get it back working, or an honest answer if it can't be saved.
Uploaded for temporary server processingSelect a PDF or drag & drop
One PDF, up to 25 MB
This tool uploads your file to our server. Repair PDF needs server-side processing, so your file is sent over HTTPS and handled using temporary server files in a separate job directory. Cleanup is attempted when processing ends, but immediate or secure deletion cannot be guaranteed. Don't use this tool for files you can't upload to a third party.
How to repair pdf
- Select your PDF. Choose or drag in the damaged file.
- Repair. Press Repair PDF. The file's internal structure is rebuilt: broken cross-reference tables and object streams are reconstructed.
- Download. Download the repaired PDF, or read the error if the damage is beyond this tool's repair capabilities.
Want more detail? Read our full guide: How to repair a damaged PDF.
Why do PDFs break?
A PDF looks like one flat document, but under the hood it's a collection of numbered objects (pages, fonts, images, streams) tied together by a cross-reference table that records the exact byte offset of each one. That table is how a reader jumps straight to page 200 without scanning everything before it. When even a few of those offsets are wrong, the reader looks in the wrong place, finds nonsense, and gives up with a "file is damaged" or "can't be opened" error, even though the real content is usually still sitting inside the file, intact.
The offsets get corrupted in ordinary, avoidable-sounding ways:
- A download or email transfer that was interrupted partway
- Losing power or force-quitting an app mid-save
- A syncing or storage glitch that flips or drops a few bytes
- A buggy exporter that wrote a slightly malformed structure
Repair works by ignoring the broken index and rebuilding it from scratch. It scans the whole file for the objects that are actually present, reconstructs the object map and a fresh cross-reference table, and writes a clean, valid structure around the content it recovered. The pages, text, images and metadata are carried over as-is, so nothing visible changes, only the plumbing that readers rely on. This resolves the large majority of "won't open" files.
What repair can't do is invent data that was never received. If a download cut off halfway, the bytes after the cut are simply gone, and no tool can rebuild text that isn't there. In that case the honest fix is to fetch or re-export the original again. This tool says so plainly rather than handing back an empty shell. This repair uses qpdf on our server, so your file is sent over HTTPS and rebuilt using temporary server files in a separate job directory. Cleanup is attempted when processing ends, but immediate or secure deletion cannot be guaranteed.
Frequently asked questions
Is my file uploaded?
Yes. This tool uses qpdf on our server. Your file is sent over HTTPS and processed using temporary server files in a separate job directory; cleanup is attempted when processing ends, but immediate or secure deletion cannot be guaranteed.
What can be repaired?
Structural damage: broken or missing cross-reference tables, damaged object streams, files that were truncated or byte-corrupted. These are the most common causes of 'couldn't open' errors, and rebuilding the structure recovers them.
What can't be repaired?
Missing content. If pages were cut off when a download was interrupted, the bytes after the cut are simply gone, and no tool can rebuild text that isn't in the file. In that case, download or re-export the original again.
Does repair change my document?
Only its internal plumbing. The pages, text, images and metadata are copied as-is into a rebuilt structure. Nothing visible changes.
The repair failed. Now what?
If the file was truncated in transit, the fix is re-downloading the original. If it's corrupted storage, restore from a backup. If neither exists, the file's contents are genuinely lost, and we say so rather than hand you an empty 'repaired' file.