Repair PDF

Rebuild a PDF whose cross-reference is broken so strict readers accept it again.

Runs in your browser. Nothing is uploaded — your input is processed on this device and discarded when you close the tab.

Your file

Drop a PDF here or click to browse. One file at a time, up to 50 MB. Everything is processed on this device only — no file is uploaded.

Repair PDF

Repair rebuilds the file structure (cross-reference and trailer) from the objects inside it. It cannot restore content that is missing, and it cannot open password-protected files — the limitations on this page explain both.

Sometimes a PDF will not open — or opens in one reader and not another, or shows an error about a cross-reference table — even though the document itself was fine when it was created. A common cause is structural, not content, damage: the file's cross-reference table (the index that tells a reader where each object lives) holds wrong byte offsets, or its startxref pointer aims at the wrong place. Truncated downloads, interrupted saves, buggy merges and some email gateways all produce exactly this injury. This tool performs the basic repair for it: it reads the document by scanning and parsing the objects themselves rather than trusting the broken index, then writes the whole thing back out with a fresh, consistent structure. The rebuilt file is the download — all in your browser, with nothing uploaded.

The honesty matters as much as the mechanism, because "repair" is a word other tools stretch. This tool rebuilds structure; it does not and cannot restore content that is not there. The limitations below state precisely which failures it addresses and which it does not, so a file that needs a different remedy is recognised quickly instead of being sent through a repair that was never going to help it.

How to use Repair PDF

1. Add the damaged PDF. Drop it onto the workspace or click to browse. One file at a time, up to 50 MB. Unlike the other tools here, this one does not need the file to open cleanly first — a broken cross-reference is exactly what it is for.
2. Read the diagnosis line. When a file is added, the tool checks its cross-reference: following the startxref pointer and verifying that table entries point at the objects they claim to. If it finds a problem, the status line says what it found in plain terms.
3. Choose "Repair PDF". The document is parsed from its objects and re-saved with a new cross-reference and trailer. Document metadata — the title and author — is carried across explicitly, and the page count that was recovered is recorded.
4. Read the result line. It reports how many pages were recovered and the before-and-after sizes. A repair that recovers fewer pages than the document should have is a partial recovery, and the count is how you know.
5. Verify the download in a strict reader. Open the repaired file in the application that rejected the original. Skim every page — especially the last one — and confirm nothing is missing before the repaired copy replaces the original in your records.
6. Keep both files until you have checked. The original is never modified, and if the repair turns out to be partial, the original is still your evidence for what the document contained.

Features

• Tolerant loading that parses the document's objects directly, so a wrong cross-reference table or startxref pointer does not block the read.
• A full structural rebuild on save: a new, self-consistent cross-reference and trailer that strict readers accept.
• A built-in diagnosis that follows the startxref pointer and verifies cross-reference entries, reporting the specific problem it finds.
• Metadata carried across the rebuild: title, author, creator and producer are preserved rather than dropped.
• A plain recovery report: pages recovered and exact byte counts in and out, so a partial recovery is visible instead of silent.
• Typed refusals: encrypted files are reported as password-protected and non-PDF files as corrupt — never dressed up as successful repairs.
• Processing runs entirely in the browser with the pdf-lib library; the original file is never changed.

Worked example

The worked example is this tool's own test scenario, run by its test suite. Start with a healthy three-page document carrying the title "Quarterly figures" and author "A. Example", saved with a classic cross-reference table — 2,707 bytes in its damaged form. The damage is applied deliberately: every offset in the cross-reference table is zeroed and the startxref pointer is reset to 0, so a reader that trusts the index looks for every object at the start of the file. The tool's strict check confirms the damage (the pointer no longer lands on the table), the repair then recovers all 3 pages, the extracted text of every page is intact, the title and author survive, and the rebuilt 1,670-byte file passes the same strict check that the damaged file failed. An already-healthy file passed through the same repair comes out with its page texts identical, page for page — the repair changes structure, not content.

Common errors and limitations

• Repair rebuilds structure only. If a page's content itself is missing or corrupt inside the file — a truncated image stream, a half-written page — the rebuilt document will faithfully carry that same damage. Nothing is invented to fill gaps.
• Password-protected files cannot be repaired open. An encrypted file is reported as password-protected and refused; this tool has no password prompt and does not remove encryption.
• A file that is not really a PDF, or a header followed by garbage with no parseable objects, is refused as corrupt. There is nothing to rebuild from.
• Annotations, bookmarks and interactive form behaviour are carried as parsed by the library; an exotic structure that the parser cannot read will fail the repair rather than be approximated. Check form-heavy documents page by page after repair.
• The repair writes a standard structure with object streams; a workflow that depends on an unusual original layout (for example incremental-update history) gets a clean single revision instead — the earlier revisions embedded in the damaged file are not preserved.
• The file cap is 50 MB and one document at a time, because the rebuild happens in your browser tab.

• Expecting repair to recover a file that was never fully downloaded. If the download stopped halfway, the missing pages' objects are simply absent, and no structural rebuild can conjure them. Fix: re-download the file; use repair when the file is complete in size but readers reject its structure.
• Assuming a repaired file is a verified file. The tool reports the pages it recovered, and a document can be structurally sound while still missing content it should have. Fix: compare the recovered page count with what you expect, and skim the result before filing it.
• Sending a password-protected file for repair. Encryption is not damage, and the tool refuses encrypted files rather than attempting them. Fix: open the file in the application that knows the password, save an unprotected copy, and repair that if it still needs it.
• Deleting the damaged original. Until the repaired copy has been opened and checked in the reader that matters, the original is the only other evidence of the document. Fix: keep both until the check is done.

Frequently asked questions

Related PDF tools

Prefer everything on one page? The original PDF Toolkit merges, splits and rearranges with one shared file list. The full suite is indexed on the PDF Tools hub.

Related articles