Moving from Mendeley to Zotero
R Shell
Switch branches/tags
Nothing to show
Clone or download

Adios Mendeley

Move from Mendeley to Zotero preserving notes, dates, folders/collections, and PDF annotations.

I have used the code here (and code from to move from Mendeley to Zotero. In this moving I haven't lost any of the Mendeley annotations for entries, the annotations of Mendeley in the PDFs themselves, Mendeley's folder structure (somewhat equivalent to Zotero's collections), or the date the reference is added.

A few other files here allow me to automatically extract annotations and highlights from PDFs (annotations and highlights in the PDF itself), as well as prepare the BibTeX file for easy usage with Emacs and with a tablet.

Update (2018-06-29): READ THIS FIRST

Zotero has now a way to import directly from Mendeley. You most likely want to try that first (and second, and third, ... before playing around with these scripts, messing around with R, etc). These are two links:

(and, for the sake of historical records, you can look at where Brenton Wiernik alerted me to the issue, and the conversation that followed).

Using it

  1. Start R, make the needed changes in the first four variables defined in mend-to-bibtex.R, and run it, paying attention to possible errors and warnings. In particular:

    • Mendeley makes no provision to avoid repeated BibTeX keys. So go and fix them: the run will give you the titles of the files with repeated keys.

    • Mendeley will not export a BibTeX entry if there is no BibTeX key. You might have deleted the key, or if you set Mendeley to automatically create keys, no key will be created if there is no author and no title. Again, the program will tell you which entries have no key.

  2. Once you have your new BibTeX file, before importing it into Zotero, open it up in JabRef. Why? Because there can be things that are not OK, like a keyword entry with newlines, or latex code itself in the notes, and this will break the import into Zotero (since this only happened to me in two entries, I did not change the code to catch and fix these issues).

If you first go through JabRef first, you will find the problem very quickly and easily (as easy as being told about the problematic line number). Fix any remaining issues.

  1. When done, import into Zotero. Beware that imports of large bibliographies can be slow, even more if you enable full PDF indexing ---I do the import first without PDF indexing, by changing the settings to 0 pages and 0 characters, and only after I have my full bibliography in Zotero I enable the indexing back.

Beware that I've only imported into a Zotero that had Better BibTeX, BBT, as an extension. I am not sure, but this might add also some extra functionality in the import (it definitely does for the export of BibTeX and I find it absolutely essential).

  1. Finally, when you import in Zotero, pay attention to the "not imported" note you might see as one of your last references. In my case, two references that imported fine in JabRef did not in Zotero (weird characters in authors' names). You will see in the text of the note something like "Better BibTeX could not import ...". Fix those entries if they are just a few, and add them by themselves (possibly adding them to their corresponding collections), or fix the entries in the complete BibTeX and go back to step 3.

  2. Fix the dates. My naive expectation that the timestamp field would be recognized does not work. But you can use Alex Seeholzer's mendeley2zotero.

    1. Make a backup copy of your databases.
    2. Modify the script to have line d_added = datetime.datetime.utcfromtimestamp(d_added) be d_added = datetime.datetime.utcfromtimestamp(d_added/1000) (at least as of July 2015 Mendeley is using time in miliseconds since 1970). (This might have been fixed by the time you read this, since I filed an issue).
    3. Run the script with the added_dates option. You might get some warnings that, as far as I can tell, are inoquous.
    4. Copy the zot.sqlite as the new zotero.sqlite, start Zotero, and check.

Enjoy! Mendeley is now just a bad dream of days gone by. :-)

Using it: what operating systems?

I've only used this with GNU Linux. I guess it should work in other Unixes (Macs, for instance) but some things might not work with Windows directly, and you'll need to modify the code.

What and how: details

I use both the BibTeX file exported from Mendeley and the sqlite database. The BibTeX file that Mendeley exports is missing many things that you probably want to keep, so I query the sqlite database, grab that info, and place it in a new BibTeX file. In the process, I fix a few major problems with things Mendeley does with files, etc. Finally, from the sqlite file I also extract the folder structure, and I add that to the new BibTeX file using the format that JabRef uses for groups. Since the Zotero import can use JabRef's groups' information (see this ZotPlus closed issue), you do not need to enter any of the folder/collection information by hand. The functions for doing all of this are defined in sqlite-bibtex-functions.R.

These are the pieces of information I get from the Mendeley sqlite file (and some details of what is done and where they are left in the BibTeX file):

  • Date the file was added (field timestamp). But note this is not directly incorporated into Zotero (see step 5 in Using it).

  • Annotations in the entry. You might think you have these in the BibTeX file, but maybe you don't: as of July 2015, with Mendeley's v. 1.14 the exporter is buggy and you will not get the full annotations in the entry if there are newlines, for instance; this is a know bug (search the Mendeley forum). So I circumvent the bug by grabbing all of the annotations from the sqlite db (and include any that might have been placed in the annote field and not be in the sqlite field text in table DocumentNotes ---nope, I did not check if this happened at all; I just took the safer way of using the union of both). These are all in the annote field in the BibTeX file, and Zotero will recognize this as notes.

  • Annotations in the PDFs made from within Mendeley. If you are naive enough (as I was when I started using Mendeley) you might be using Mendeley's PDF editor which does/did NOT place the annotations and highlights in the PDF itself (this is easy to check if you open a Mendeley-annotated file in any other pdf viewer). So I grab the PDF annotations and place them in a new field in the BibTeX file that I call mendpfnotes. You might, alternatively, want to use Menextract2pdf that "extracts highlights and notes from the Mendeley database and adds them directly to all relevant PDF files, which can then be read by most PDF readers".

  • Keywords and tags. In Mendeley's dbs and BibTeX you get a keywords and a mendeley-tags field. I think keywords often contain the author's keywords, but not always, and one of the two can be missing or overlapping the other. So I return a single field (called keywords), which is the union (after tokenizing by commas) of keywords and mendeley-tags. This is happily recognized by Zotero as tags.

  • The original Mendeley id, in case you need to go back to the sqlite database (field mendid).

Some clean up operations

I also do some cleaning up of strings, such as some of the HTML markup in the notes that will prevent proper import and/or make life harder when reading them (e.g., I think you definitely want a newline if you have a <br/>).

Another important clean up operation affects file names. Again, if you allow Mendeley to rename files for you (which was a silly thing I did sometime in the past) you will find that you end up with file names that can be really long and that contain spaces and other annoying characters. I fix this before importing into Zotero: any file with a name longer than a user-specified length (I used 40, but this is a user-modfiable function argument), or with spaces or characters that are not letters, numbers, the period or the hyphen, are renamed using as the new name the BibTeX key and a randomly generated sequence of letters and numbers (length of the random part is also a user-modifiable argument).

I actually wonder why Mendeley can create such long file names. This seems like an obviously bad idea (I once tried to encrypt a home partition using eCryptfs or encfs, or both, and hit the maximum length possible, which is not difficult, given that Mendeley can also create horrendously long directory names).

Caveats and warnings

  • Getting the collections/folders. I've only used it with a set of folders (Mendeley jargon) or collections (Zotero jargon) of up to two levels of depth. I think the code should work with arbitrarily deeply nested structures, since the algorithms I wrote are fairly simple, but I have not checked it.
  • If you have latex code in the notes, it can break things (it did for me) during the import.
  • If you have keywords that contain things that get interpreted as newlines in the BibTeX text file, it can break things during the import.
  • I have noticed that some of my lower-level collections have been duplicated as upper level. I think this is a different issue that appeared later as I did something wrong. But check it. (I say this because JabRef does not show any problems). Anyway, this should be a matter of removing the duplicated upper-level collection (not the collection and items).

Why through BibTeX

Initially, it looked like a good idea:

  • Because to me it is much simpler than trying to directly manipulate the Zotero sqlite database.
  • You can incorporate loads of additional information, including the Mendely folders the tags, etc, as new BibTeX custom fields.

However, maybe I should not have used the originally exported Mendeley BibTeX file and modified it, but just directly created the BibTeX file from scratch from the sqlite db.

Why R?

The initial getting the info out of the Mendeley db involves just a couple of queries to the sqlite database which I do using the RSQlite package. The rest is massaging and cleaning, checking, modifying the bibtex, and preparing the JabRef group structure. I do it from R just because it is easier for me than using, say, Python.

Other ways of doing this

Ideas and suggestions of how to go about moving from Mendeley to Zotero are available in Zotero's forum. For instance this and this. But most of the ideas still involve a fair amount of manual work. Alex Seeholzer has written a Python script to move from Mendeley to Zotero but it did not fully fit my needs; in particular, the folder/collection structure needs to be created before hand in Zotero, I think the PDF annotations by Mendeley do not get transferred, and I think you might not get the full annote and keywords fields. As you have seen, though, I use it for fixing the dates (see step 5 in Using it). Menextract2pdf "extracts highlights and notes from the Mendeley database and adds them directly to all relevant PDF files, which can then be read by most PDF readers".

Why move away from Mendeley?

There are several good reasons:

  • Mendeley is not free software. It is not even open source.

  • It is not possible (or I have not been able to) just recover all of your database from a backup you have yourself. I do not mean the backup that Mendeley creates, I mean the backups you create by whatever reasonable backup procedure you use, which creates copies at specific time points, etc. If you have a decent backup policy of your machines and on day t your db gets screwed up, you might think that it is just a matter of going to your backup of t - k, and restoring from there (maybe you are lucky and k is less than a day old). Well, nope, it does not work that way and there is no simple way to say "here, this is the database; use this, and forget anything and everything in your servers". (And yes, I asked about this in the Mendeley forum to no avail.)

    With Zotero, in my limited tests, it does: in desperate cases you can tell it to overwrite all the server stuff with your database (the "Restore to Zotero servers"). And if you keep a backup of your BibTeX files, you might also be able to restore most or even all of your stuff from there (at least if you use BBT and export notes, etc).

  • Mendeley does not check for duplicated BibTeX citation keys and will not generate a BibTeX entry if you do not have author and date entries. This sucks, as for me the place I look for stuff is the BibTeX file (much faster searching than via Mendeley itself).

  • Mendeley, in my opinion, has a cluttered interface, compared to Zotero.

  • Mendeley, in my experience, is much worse at getting the reference information from the PDF itself and will insist in screwing up when you ask it to review a reference or use the DOI. Too much manual intervention.

  • Mendeley's searchers are slow. Zotero's are not blazing fast, but I find them faster.

  • Sorting files into subfolders: who thought it was a good idea to use spaces in the "folder" (directory) name? What about spaces in the file name itself? What about apostrophes, or question marks, or ...?

  • In general, it is easy to shoot yourself in the foot (or it is for me).

  • Keyboard shortcuts: few, and where are they documented? I find it unavoidable to use/abuse the mouse (I often end up with wrist pain after intensive bibliography work with Mendeley).

  • Inflexible about what fields you are shown (some you cannot uncheck, which sucks with long abstracts).

  • Persistent errors in entries, like adding links to files that you have removed which results in the broken link in Mendeley and paths to inexistent files in the BibTeX file.

Why not before?

Because of the manual or programming work involved, I did not decide to take the plunge until a recent Mendeley crash that did not allow me to recover my library to a sane state. As well, I have been very happy with my usage of Referey to read papers in a tablet (see my former web page entry).

Using a tablet

This is left here for historical purposes, but it is OUTDATED!!! I am now using Referey, after converting the Zotero db to a db that Referey will understand. The code is available from the repo Zotero-to-Referey.

Here things are not nearly as easy and nice as they were with Referey and Mendeley. I have tried all of the apps listed in (, including Zandy, Zed Lite, Zed beta, Zojo, Zotero Reader, Zotable, and Zotfile. The main problem is that with none of them you can do something that to me is necessary: get the database file from whererever the app deams suitable, but take the PDFs (more generally, attached files) from a place I sync on my own. I do not store my attached files in Zotero's servers and I do not use Dropbox, but sync them with syncthing. I do not want to set up a WebDAV server either, since it is just simpler to keep my Zotero storage directory (where the PDFs live) synced between computers and tablets with syncthing. Zotfile is nice, but not really what I need and I find it too complicated and requires too much manual intervention ---see the comments in my web page entry; for instance, I need to decide what to send back and forth; too much work, and requires making accurate predictions about what I'll want/need to read.

So what I am doing now is using Android applications that understand BibTeX files. There are three: RefMaster, Library, and Erathostenes.

Erathostenes understands the nesting in JabRef's groups, though it will only display the lower-most level, if you have several levels of nesting. So you can see part of the structure of your Zotero collections. However, Erathostenes takes a very long time to load my BibTeX file. It is Erathostenes what I use most of the time, but I am not fully satisfied. (Erathostenes might support Zotero in the near future: see the changelog for future versions).

RefMaster is nice (sorting by several fields) and in my experience faster than Erathostenes, but it does not support multiple files per entry (it is a "future feature", but this might take long to arrive, given that RefMaster's last version is from over two years ago). JabRef groups are not supported yet either.

Library is much faster than the two above and supports multiple files and you can of course search the library. However, you cannot sort by date (this is a requested feature) and there is no support for JabRef's groups.

Things, thus, are not as easy and simple as they were with Referey, and I do miss that.

How do I use Zotero and the tablet?

Again, this is left here for historical purposes, but it is OUTDATED!!! I am now using Referey, after converting the Zotero db to a db that Referey will understand. The code is available from the repo Zotero-to-Referey.

First, I only use the tablet to read and annotate the PDFs. I am not that interested now in modifying the BibTeX file (or Zotero's db) itself. So this is just a matter of getting the BibTeX exported from Zotero into the tablet. I use syncthing for syncing the PDFs and other attachments and for syncing the BibTeX file.

But what BibTeX file? Not the one immediately exported by Zotero, since that has stuff I might not need (the collection structure if using Library or RefMaster) and it is better to change the file paths. So I process Zotero's BibTeX with a little sed script that I run whenever Zotero produces a new BibTeX file. See all the details in Automatically propagating changes in the database to helm and tablet.

Odds and ends

Extracting all PDF annotations and placing them in an org-mode file

Beware: the approach explained here is not working well. I find that I am missing many annotations or highlights made in the tablet with at least one PDF reading application. The annotations are there, you can see them in Emacs with pdf-tools or in Okular or whatever, but neither leela nor the ruby script will extract them. Fortunately, Zotfile can extract them. If you do not want to miss annotations/highlight, use Zotfile from Zotero and extract annotations. Then, run and advanced search searching for your term in "Note" (not "Annotation"). Thus, what follows is likely useless.

A very simple way, for me, to be able to quickly search all annotations in PDFs (and even highlights) is to extract the annotations and highlights, and place them in an org mode file, where the heading is an org mode link to the file and for each file I have all the comments and highlights. Then, I can easily search from them. To do that, I have a cron job that every night calls Rscript with leela-ruby-extract.R.

I am using Leela (see my old web entry for details) and also this ruby script by Dan Lucraft that uses pdf reader. The ruby script will extract highlights too, whereas that is not working with Leela.

It seems that pdf.js is a very capable platform that extracts highlights and annotations, and that is in fact used by Zotfile. But I have no idea if/how to run it from the command line as just a simple standalone, although there are pointers out there I could not figure out how to follow quickly.

Zotfile, a Zotero extension, allows for similar things (or even better, depending on your point of view) but it requires manual triggering (and, file by file, is a lot slower than Leela). With the approach I use, extraction takes place automagically.

Finally, it is also possible to do this directly from Emacs itself with pdf-tools, and include the notes in an org file: Note Taking with PDF Tools. It works great, and not only do we get both notes and the text of the highlight (i.e., what I was doing through the ruby script or Zotfile), but we also get a link to the precise location of the annotation thanks to the added "++" syntax in org-pdfview. However, I still have to figure out how to, automatically from a cron job, generate a single file with all annotations (or scan my library to generate one file of annotations per PDF).

So there is much room for improvement here:

  • Trigger the extraction only for the PDF that is modified (and neatly insert the annotation in the proper place).
  • Populate the annotation file only for files with annotations (not with useless links to PDFs that contain no annotations).
  • Create the annotation file in the best way for helm-bibtex to understand (helm-bibtex supports keeping all notes in one file: see this commit and this news entry).
  • Use the approach in Note Taking with PDF Tools to generate that file for helm-bibtex.
  • Figure out whether I want a single file with all annotations or a file per entry (preliminary tests suggest that a single huge file will lead to deals in helm-bibtex).
  • Add the notes from Zotero itself to the org file(s) with annotations, or merge them into just one annote field. This way, I'd have a single place to do all and every search (helm-bibtex can search on additional search fields, but if you have several per entry only the last one is used).

Using the BibTeX file from Emacs

Left here for historical purposes, but now OUTDATED helm-bibtex can, since mid November 2015, deal with the File field (see this merge and this commit). So I am still using helm-bibtex, of course, but I no longer need to do any of the things below.

After seeing it mentioned in the org mode list, I've started using the really great helm-bibtex, a bibliography manager for Emacs. Searching for stuff and inserting references in something you are writing is amazingly simple and powerful. However, I am not using helm-bibtex-find-pdf since I often have multiple PDFs associated with one entry. (I also have other issues, such as how the path is specified, but these could be fixed with the hints that Titus von der Malsburg gave me here).

What I do, if I want access to the PDF, is open the entry I want from helm-bibtex and then go the the file field, and C-x C-f to open the file (I am using ffap with ido, so C-x C-f on top of the file path opens it). I am now using pdf-tools to view the PDFs from within Emacs, but it would work the same with Okular or another viewer.

In fact, it is very simple to have helm-bibtex jump to the file field directly as the default action (and, once in there, if you have multiple files, moving between them is simple with forward-sexp). I have this in my .emacs:

(defun rdu-helm-bibtex-go-to-file-field (KEY)
  "Jump to the file field in the entry."
  (helm-bibtex-show-entry KEY)
  (search-forward "file = { ")

(helm-add-action-to-source "Go to file field" 'rdu-helm-bibtex-go-to-file-field
			   helm-source-bibtex 0)
(helm-delete-action-from-source "Show entry" helm-source-bibtex)
(helm-add-action-to-source "Show entry" 'helm-bibtex-show-entry
			   helm-source-bibtex 1)

For that to work, I generate, from the Zotero BibTeX file, a bib file with the file paths stripped of extraneous information, so that the file field contains only file paths. This is done with the script, and further details are provided in Automatically propagating changes in the database to helm and tablet.

Beware that because of the way BBT works, if you want to preserve the full path, you might not want to export to a bib file that lives right in your data directory. See this thread. (So, in my case, I export directly to ~/Zotero-storage/zotero-$HOSTNAME.bib, not to ~/Zotero-data/storage/zotero-$HOSTNAME.bib). You mileage might vary and, regardless, this would be an issue of modifying the sed script to your needs.

Automatically propagating changes in the database to helm and tablet

Outdated too: not needed any more. See above and Zotero-to-Referey

Whenever there is a change in the Zotero database, the BiBTeX file gets updated (this is something you configure in Zotero, and I have it so that each machine running Zotero writes its own zotero-$HOSTNAME.bib file). But we want to modify this file, so it is easier to use in the tablets and with Emacs. We want BiBTeX files without the JabRef group structure for Library and RefMaster (we leave it if using Erathostenes) and with easier to use file paths (easier to open from Emacs and the tablets). This I do with the sed scripts

Since we want to have these changes propagate automagically to the tablet and the file I use with helm-bibtex, all that remains to be done is run the script when the BibTeX file changes. We could do it with inotifywatch but using entr is much simpler. So that I do not need to remember to launch it manually, in my .xsession I have

ls ~/Zotero-data/storage/zotero-$HOSTNAME.bib | entr ~/Adios_Mendeley/ &

The reason why I have specific files per host is explained in the Notes about using syncthing section

Notes about using syncthing

I use syncthing for syncing the PDFs and other attachments. Of course, beware that you most likely do not want to sync the sqlite files themselves via syncthing (read the Zotero docs and forum). My actual directory structure is to keep the Zotero db under ~/Zotero-data and have the storage subdirectory be a symbolic link to ~/Zotero-storage. All the attachments and the BibTeX files live under ~/Zotero-storage and that gets synced (except for the files mentioned below).

Why do I have different bib files in different computers? The bib file is exported, from Zotero, to the storage directory, but since any change in the bib file triggers a change in the helm and tablet bib files, I do not want to have this triggered in all the computers almost simultaneously if they are online: this could lead to conflicts via syncthing (the helm and tablet bibs being changed about the same time in all online machines). By keeping different Zotero bib files per machine, the bib in the machine I am working might change, and the changes in the helm and tablet bib will only appear in the machine I am working, and then be propagated to the other machines without leading to conflicts.

If you use PDF full text indexing, in the storage directory you might want to add an .stignore file as follows:


to prevent the indexes from being sent to the tablet (where they are probably of little use).


All the code here is copyright, 2015, Ramon Diaz-Uriarte, and is licensed under the GNU Affero GPL Version 3 License.