How to Merge and Split PDFs in Seconds — No Sign-Up, No Watermark
Reviewed by the FreeOnline.fyi team · Updated 2026-09-15
A PDF workbench, not a PDF suite
We built Custom PRD build around two jobs and nothing else: merge 2–20 PDFs into one file, or pull a page range out of a single PDF and hand it back as a new document. No account, no email loop, no watermark stamped across your signed contract, no upsell screen between you and the download button.
The processing is real, and it is deliberately boring. Pages are copied from the source documents into the output using pypdf, which manipulates the PDF's internal object structure rather than its pixels. That distinction matters in practice: nothing is rasterised, nothing is re-compressed, so a 300 dpi scan goes in at 300 dpi and comes out at 300 dpi, and typed text stays selectable.
The trade-off is that this is a structural tool, not an imaging tool. There is no OCR, no deskewing, no cleanup of a crooked phone photo. If you want the technical background on why PDFs behave this way, the format is defined in the ISO 32000-1 standard and summarised in the Wikipedia entry on PDF.
Merging: the five-step run
1. Open the tool and leave the toggle on Merge. 2. Drop in 2–20 PDFs, up to 100MB total. 3. Check the file summary list on the right — that list is the order your pages will appear in. 4. Type an output filename if you do not want the default (path characters get stripped and the .pdf suffix is forced, capped at 60 characters). 5. Hit run, then use the download card.
A concrete example from our own testing: five invoice PDFs at 3, 2, 8, 1 and 4 pages should produce an 18-page merged file. Before you download anything, read the page count on the result card and do that arithmetic. If the number is wrong, the problem is almost always a file you forgot was in the queue, not the merge itself.
Order is the step people skip. If the summary list shows the signed contract before the cover letter, fix that before running — reordering merges afterwards means starting over. The tool also does not rebuild outlines, bookmarks or page numbers, so a merged file of ten chapters will not arrive with a tidy table of contents.
Splitting: ranges, every-page mode, and the ZIP caveat
Split mode takes exactly one file. You then type a page range in the syntax 1-3,5,8-10, which is read as a list of individual pages and closed ranges. Ranges are validated against the real page count, inverted ranges like 7-3 are rejected, and if a page number does not exist you get an error naming the problem rather than a silently short file.
There is a second switch that catches people out. In extract-one mode you get a single PDF containing just the pages you asked for — useful when you want pages 3–7 of a 200-page report to email to a colleague. In every-page mode the page range is ignored entirely and you get a ZIP archive with one PDF per page, so a 40-page scan batch becomes 40 separate files. If your download arrives as a .zip, that is expected in this mode, not a bug.
Both modes suit different jobs. Pulling a signed signature page, an appendix or a single invoice out of a bundle is extract-one work. Splitting a scanned stack of forms so each person gets only their own page is every-page work — just budget a minute for renaming afterwards, since the tool names files by page.
Reading the result card and the error banner
The download card gives you the three numbers worth checking: output filename, total page count, and final file size. The summary list underneath shows what went in — per-file name, page count and size for a merge; per-output page numbers for a split.
Do not be alarmed if a merged file is not exactly the sum of its inputs. PDFs frequently share fonts, colour profiles and image resources, and the library can reference those once instead of copying them twenty times, so the output can be noticeably smaller. It can also grow slightly when documents use different resource sets. Neither direction means loss.
When something fails, the error banner tries to be specific: which file, which page, what was wrong. That is intentional. Most tools in this category say "something went wrong" and leave you guessing; we would rather tell you that file three is not a PDF or that page 412 does not exist in a 200-page document.
Mistakes we see, and the honest limitations
Renaming a .docx or .jpg to .pdf is the most common failure. The upload is checked against the file's magic bytes, not its extension, so the mismatch is caught immediately. Encrypted or password-protected PDFs are the second — if a file needs a password to open, expect the run to fail rather than quietly succeeding.
The 100MB total ceiling is a real constraint on high-resolution scans. A 200-page colour scan at 600 dpi can blow past it on its own, and no amount of merging will help. Check the combined size in the summary list before running, and split the job into two batches if needed.
Finally, remember what is not here: no OCR, so a scanned document stays an image of text; no page reordering across files beyond the upload order; no compression, cropping or redaction. For anything that affects a legal or financial document, verify the output page-by-page before you send it — a tool that does two things well still cannot check your work for you.
Why there is no sign-up, and what happens to your files
The tool is stateless by design. There is no account, no storage bucket holding your documents for a week, and files are removed after the run completes. That design is what FreeOnline.fyi is built around generally: open the page, get the result, close the tab.
Treat "files deleted after processing" as a reasonable engineering claim from us, not as an audit certificate. If your organisation has a policy about where client documents may be uploaded, follow the policy — for genuinely sensitive material, a local desktop PDF utility is the safer call, and we would rather say that plainly than pretend otherwise.
The same no-frills approach shows up in our other small tools, from Photo to Maze Generator if you want something playful to Image to ASCII Art when you need a terminal-style version of a logo. Different jobs, same rule: the tool does one thing, tells you exactly what it produced, and gets out of the way.