MirrorDL

Download files from several mirrors at once, after checking that they all serve the same file.

Paste the links to one file or to many: MirrorDL sorts them into files, shows you what it found, and downloads each file from all of its mirrors at once. It can also continue unfinished downloads that Firefox or aria2 left behind. It runs on Windows 10 and 11, as a window app and as a terminal program.

Features

• Many files, sorted by themselves. Paste links to several files, also several mirrors of each: links with the same name, or the same size and content, are mirrors of one file; the others are files of their own. Before anything is downloaded, you see the files - name, size, links, what is wrong with a link - and tick what you want.

• A download list. Start new downloads while others run; pause, resume or take off each one, or all at once. The list is kept: after closing MirrorDL or restarting the PC, press Resume.

• Links from web pages. A link to a file host's page is searched for the link behind its download button. A page that needs a captcha opens in MirrorDL's own browser: solve it, click the download button, and MirrorDL takes the link.

• History and settings. Finished downloads are listed, searchable, with their links. Every extra - the history, keeping the list, watching the clipboard, keeping the PC awake, the taskbar progress - is a switch in Settings, with what it does. The theme follows Windows, or is light, dark or OLED black (a pure black window).

• One window. Starting MirrorDL again, or dropping files on MirrorDL.exe, hands them to the window that is open.

• All links at once. Each link gets several connections. Faster links keep taking over work from slower ones until the end.

• Finds the best number of connections by itself. It adds connections to a link as long as they make the whole download faster, and stops where the server or your internet line is the limit, at most 16 per link. Nothing to set.

• Checks the links first. Links with a different size are skipped. MirrorDL also compares 64 KB pieces from several places in the file across all links and skips links whose content differs, so you never get a file stitched together from two different versions.

• Continues unfinished downloads, also ones from other programs:

  • Firefox .part files, including Firefox's random names (setup.Ab3-xY_z.exe.part)
  • aria2 downloads: reads the .aria2 control file to find the missing pieces
  • its own downloads, even after a crash or a power cut
  • several unfinished files of the same download at once, also under other names: it puts together what they hold and downloads only the rest

  Before continuing, it compares what is already on disk with the links. A file that doesn't match is never changed. Continuing reads only the progress file and a few pieces, not the whole file, so even a 50 GB download goes on right away.

• Verifies the result against a checksum (MD5, SHA-1, SHA-256, SHA-384 or SHA-512) if you paste one with the links - also several, for several files: with the file's name next to it (as in a SHA256SUMS file), or without, and MirrorDL finds which file it belongs to.

• Shows the details: percent, downloaded and remaining size, current and average speed, time spent and left, connections, a line per link, and a map of which parts of the file are done.

• Copes with real servers: retries dropped connections, uses fewer connections on servers that allow only a few, and saves the progress every 10 seconds and when you pause or close the window.

Download

The window app from the Microsoft Store, which installs it and keeps it up to date:

or in a terminal: winget install 9P94PN403J5W

Or from the latest release: MirrorDL-win-x64.zip is the window app, mirrordl-terminal-win-x64.zip the terminal program. Unzip and start; nothing gets installed.

They need Windows 10 (1809 or newer) or Windows 11, 64-bit, and the .NET 10 Runtime. The window app brings the Windows App SDK along.

Updates: the Store version gets them from the Store. The window app from the zip tells you when a newer release is out, with a link to it. It asks GitHub once at start and sends nothing else (privacy policy).

Use

Window app

Start MirrorDL.exe, paste the links, and press Check links. MirrorDL shows the files they lead to; tick the ones you want and press Download.

• Every link, checksum and unfinished file has a box of its own. Paste a whole list into any box, or drop it on the window: each line goes into a box of its own, in the list it belongs to, and next to each box MirrorDL says what it made of it.
• Links kept in a file: drop a .txt or .md (.markdown) file on the window, or press From a file… next to Links (that one reads any text file up to 4 MB). MirrorDL takes the links and checksums written in it - also markdown's [name](link), lists and tables - and leaves the words, pictures and paths around them.
• The Browser button (next to From a file…) opens MirrorDL's browser to browse anywhere: a download clicked there is handed to MirrorDL and checked; it starts at once when Show the files before downloading is switched off in Settings.
• The files of one check download one after another, each with all of the internet line, so the first ones are done soonest. Check more links while they run: those start next to them.
• Each download has its buttons: pause, resume, show in folder, take off the list. Click one to see its details below the list: speed, time left, connections, its links, a map of the file, and its messages. Pause saves the progress; so does closing the window during a download.
• A file that is in the save folder already, complete, is only checked, not downloaded again.
• When all downloads are done: next to the download-list buttons, the PC can be told to sleep, hibernate or shut down once nothing in the list runs anymore. A 60-second countdown comes first that you can cancel; the choice isn't kept between starts.
• History lists the finished downloads; Settings (the gear) has a switch for every extra.

MirrorDL.exe --start <links…> fills in the links and starts at once, without showing the files first.

Links from web pages and link sites

Some links lead to a page with a download button rather than to the file. MirrorDL reads the page and takes the links to files on it. A page whose link only appears after a captcha or a click shows none: Open page opens it in MirrorDL's browser. Solve the captcha, click the download button, and MirrorDL takes the link and shows the file - the page's own download doesn't happen. MirrorDL never solves a captcha for you.

Terminal

mirrordl                                      asks you to paste the links
mirrordl "https://a/file.iso" "https://b/file.iso" -o D:\Downloads
mirrordl links.txt                            links to one file or many, and checksums (or links.md)
mirrordl "C:\Users\me\Downloads\file.Ab12-xYz.iso.part" "https://a/file.iso"

With links to several files, it lists the files it found, downloads them one after another, and sums up how each went.

Option	Meaning
-o <folder>	where to save (default: Downloads, or next to the unfinished file)
-c <n>	at most n connections in total, 1 to 64 (default: no limit)
-m <n>	at most n connections per link, 1 to 64 (default: 16)
--delete-leftovers	delete leftovers that can't go to the Recycle Bin for good, without asking
--one-file	all links are mirrors of one file: don't sort them into files
--no-pages	don't look for links to files on pasted web pages

Ctrl+C pauses. Run the same command again to continue.

How links are sorted into files

1. MirrorDL asks every link for its file's name and size, at most two at a time per server.
2. Links with the same file name - from the server, or a file name in the address - are mirrors of one file. Of those, a link with another size or content serves an old or wrong file: it is skipped, as before.
3. Links with other names but the same size are compared by a few 64 KB pieces of content: the same content means the same file under another name (a mirror that renamed it); other content means another file, as with the parts of a split archive, which all have one size.
4. A link that doesn't work goes with the file its address names, or shows as a file of its own that can't be downloaded. Two different files of the same name get numbered names (download, download (1)), also across downloads that run at the same time.

Checksums go with the file named on their line (<checksum> name, or SHA256 (name) = <checksum>) or the link on their line; with one file, all are that file's. A checksum that says nothing of its file is compared with each file at the end: the file that matches it is checked; one that matches none of them is said, but not called damaged - the checksum may be another file's. Unfinished files you add go with the file of their name, or of the size their progress file tells.

Continuing downloads from Firefox and aria2

MirrorDL looks in the save folder for an unfinished download of the same file. You can also choose one yourself.

• Firefox saves name.Ab12-xYz.ext.part and an empty name.ext next to it. MirrorDL continues the .part file and puts the finished file in place of the empty one. Pause the download in Firefox or close Firefox first. Don't cancel it in Firefox: that deletes the .part file.
• aria2 writes pieces all over the file, so the file size says nothing about how much is done. MirrorDL reads the .aria2 file to find the missing pieces. Stop aria2 first.

A file you choose keeps its name and folder, also when the links give another name. Only a .part file gets its final name once it is done.

If the file you choose is complete already, MirrorDL only checks it: against the links, and against the checksum if you paste one. It doesn't write to it, rename it or move it. That makes it a quick way to check a finished download. If the links can't send parts of the file and you paste no checksum, only its size can be checked, and MirrorDL says so.

Several unfinished files of the same download

Browsers download a file from the start to the end, so of two Firefox .part files of the same download the bigger one already holds everything the smaller one has. aria2 and MirrorDL download pieces all over the file, so their unfinished files can hold different parts. MirrorDL uses them all:

1. It finds the unfinished files of the download in the save folder, also numbered copies like name(1).ext.part. Files under other names you add yourself (several at once: one per line in Continue, or drop them on the window).
2. It compares each one with the links. A different file is left as it is.
3. The one holding the most is continued; what the others hold beyond it is copied into it. Copying on the disk is much faster than downloading.
4. Only the rest is downloaded. Once the finished file checks out, the other unfinished files go to the Recycle Bin, with their progress files.

What "checks out" means: the download finished with the right size, and the file matches the checksum you pasted, or, without one, the links at every piece compared before. A file is only removed if it was found to be this download, hasn't changed since, and no other program has it open. If anything fails, a download stops, is paused or comes out damaged, nothing is removed.

Only unfinished downloads are removed: .part files, and files with a .aria2 or .mirrordl progress file. Any other file you added stays, since it may be one of your own.

MirrorDL moves them to the Recycle Bin. A file that can't go there - too big for it (the Recycle Bin of a drive holds a few percent of it at first), or its drive has none - can only be deleted for good, so MirrorDL asks once the download is done, and keeps it if you say no. Tick Delete leftovers that can't go to the Recycle Bin, without asking (in the terminal: --delete-leftovers) to have them deleted without the question. Either way, a leftover is checked again right before: one that changed or that another program has open stays. (Should Windows delete a file for good while moving it to the Recycle Bin, MirrorDL says so.) To have big leftovers go to the Recycle Bin instead, make it bigger: right-click it, Properties, Custom size.

A file from aria2 or MirrorDL needs its progress file (.aria2 or .mirrordl) next to it, since only that tells which parts are downloaded. Without it, MirrorDL doesn't use the file.

Limitations

• Direct download links only (HTTP and HTTPS). Pages that need a login, cookies or a captcha before the download starts won't work.
• The servers must support range requests to use more than one connection and to resume. For servers that don't, MirrorDL falls back to one connection.

Build

Needs the .NET 10 SDK. Open MirrorDL.slnx in Visual Studio, or:

.\build.ps1

This builds both programs into publish\app and publish\terminal and zips them.

In VS Code (with the recommended extensions it offers): Ctrl+Shift+B builds, F5 starts the window app or the terminal program in the debugger, and Terminal > Run Task has the unit tests, the end-to-end test, the release zips and the Store package.

Project	What it is
MirrorDL.Core	the downloader: link checks, segmented download, resume formats
MirrorDL.App	the window app (WinUI 3)
MirrorDL	the terminal program
tests/MirrorDL.Tests	unit tests

Tests

dotnet test --project tests/MirrorDL.Tests/MirrorDL.Tests.csproj
pwsh tools/e2e-test.ps1 -Exe MirrorDL/bin/Debug/net10.0/win-x64/mirrordl.exe

The unit tests run whole downloads against web servers in memory (FakeServer): new downloads, continuing MirrorDL's, Firefox's and aria2's unfinished files, putting several together, checking finished files, and links that fail in every way - in about 15 seconds, without network or ports. A file "moved to the Recycle Bin" goes to a folder of the test instead. To see how much of the code they reach:

dotnet test --project tests/MirrorDL.Tests/MirrorDL.Tests.csproj --coverage --coverage-output-format cobertura

The end-to-end test starts local test mirrors (tools/testserver.py, which needs Python) that are slow, drop connections, limit connections, serve another file or ignore range requests, downloads from them with the real program, and checks every result by its SHA-256. It also moves leftovers to the real Recycle Bin. With aria2 installed, it also continues a download that aria2 started.

Speed

pwsh tools/speed-test.ps1 -Exe MirrorDL/bin/Release/net10.0/win-x64/mirrordl.exe

downloads in scenarios like a host that allows few connections, a link that slows down after a fast start or a far server, several times each, and compares the times - all on this PC, without the internet line. What was measured, what changed and which ideas didn't help is in docs/speed.md. A change meant to make downloads faster is measured with it first.

Releases

Releases are automatic. Once CI has passed on a push to main, the commit messages since the last release decide, written as Conventional Commits: fix: makes a patch release, feat: a minor one, feat!: a major one, and docs:, test:, ci: and the like release nothing. Neither does any commit that changes nothing that goes into the program (docs, tests, tools, workflows). The release gets the zips, and the Microsoft Store gets the same version (how that is set up). Each release's page shows where it is, GitHub and the Store, and a failing step opens an issue (more).

License

Copyright (C) 2026 Nawid Salehie

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. See LICENSE.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Additional permission under GNU GPL version 3 section 7: If you modify this program, or any covered work, by linking or combining it with the Windows App SDK (or a modified version of that library), containing parts covered by the terms of the Microsoft Software License Terms for the Windows App SDK, the licensors of this program grant you additional permission to convey the resulting work.