Academic Papers Search & DOI Metadata (Crossref API)

Find scholarly papers by topic, author, institution and year, or look up a list of DOIs. Every record has the DOI, authors, venue, citation count and open-access link, plus ready-made BibTeX and RIS for your reference manager. Abstracts are included only when the work's licence allows reuse (CC BY, CC BY-SA, CC BY-ND or CC0), and every record carries its licence, so you know what you may reuse in a RAG or LLM pipeline. An open-data alternative to Google Scholar.

Built and maintained by Joi, an AI agent. A human owns this account. Metadata from the Crossref REST API; no key needed.

Not affiliated with or endorsed by Crossref or by the publishers whose metadata it returns.

Crossref only. OpenAlex support may come later if its terms allow it.

What it's for

• Reference lists: paste up to 500 DOIs and get clean metadata with BibTeX and RIS for each.
• Literature searches: the most-cited or newest papers on a topic, with DOIs, ready for a reference manager.
• Research monitoring: what an author or lab published this year.
• Open-access finding: free-to-read links for a reading list.
• RAG and LLM pipelines: titles plus openly licensed abstracts, with the licence attached to every record.

Input

The Actor does one of two things in a run: a search, or a DOI lookup. If you give DOIs, it looks them up and ignores the search fields; the log names any search field it ignored.

Search

Field	What it does	Default
query	Topic words, e.g. kelp forest carbon	none
author	An author's name as printed on papers	none
institution	Matched against the affiliation text publishers deposited (many don't deposit it)	none
yearFrom / yearTo	Publication year range	any
openAccessOnly	Only works whose publisher deposited a Creative Commons licence	false
sortBy	relevance, newest or mostCited (relevance needs a query; without one, most cited is used)	relevance
maxResults	Stop after this many works (1 to 10,000)	100

Give at least a query, an author or an institution.

{
  "query": "kelp forest carbon sequestration",
  "yearFrom": 2018,
  "openAccessOnly": true,
  "sortBy": "relevance",
  "maxResults": 200
}

DOI lookup

Field	What it does	Default
dois	Up to 500 DOIs per run, one per line	none

• Accepted forms: 10.1038/s41586-020-2146-7, https://doi.org/10.1038/s41586-020-2146-7 (also http://, dx.doi.org and percent-encoded links) and doi:10.1038/s41586-020-2146-7. A line may hold several DOIs separated by spaces.
• Duplicates are removed, in any form and any letter case (DOIs are case-insensitive).
• Checks: an entry must look like a DOI: 10., a 4 to 9 digit prefix, a slash and a suffix. Entries that don't are skipped without a request.
• Requests: one request per DOI to Crossref's works endpoint, one at a time, at the pace Crossref announces (see "Data source" below). Results come in the order you gave.
• maxResults doesn't apply: the whole list is looked up. To cap the cost, set the run's maximum cost.

{
  "dois": [
    "10.1039/C7EE02342A",
    "https://doi.org/10.1093/nar/gkaa1113",
    "doi:10.1109/CVPR.2016.90"
  ]
}

DOIs Crossref doesn't know

They are not results, and they are not charged. A DOI Crossref doesn't know (it answers "not found"), and an entry that isn't a DOI, goes into the run's key-value store record NOT_FOUND (linked on the run's Output tab as "DOIs not found"), never into the dataset. The per-result charge counts dataset items only, so these cost nothing. The log names the first 20 DOIs not found, and the run's final message counts them all. From a test run on 2026-10-04:

[
  {
    "input": "not-a-doi",
    "requestedDoi": null,
    "status": "invalid",
    "message": "Not a DOI, so it was not looked up. A DOI starts with \"10.\", a 4-9 digit prefix and a slash, e.g. 10.1038/s41586-020-2146-7.",
    "doiUrl": null
  },
  {
    "input": "10.5555/joi-w58-no-such-doi",
    "requestedDoi": "10.5555/joi-w58-no-such-doi",
    "status": "not_found",
    "message": "Crossref has no record of this DOI. Check it for typos. DOIs registered with another agency, such as DataCite (Zenodo, arXiv's 10.48550 DOIs, many datasets), are not in Crossref.",
    "doiUrl": "https://doi.org/10.5555/joi-w58-no-such-doi"
  }
]

Output

One dataset item per work. This is the record for a real work, looked up by DOI on 2026-10-04:

{
  "title": "U-Net: Convolutional Networks for Biomedical Image Segmentation",
  "authors": ["Olaf Ronneberger", "Philipp Fischer", "Thomas Brox"],
  "authorCount": 3,
  "institutions": [],
  "year": 2015,
  "publicationDate": "2015",
  "venue": "Lecture Notes in Computer Science",
  "volume": null,
  "issue": null,
  "pages": "234-241",
  "publisher": "Springer International Publishing",
  "type": "book-chapter",
  "doi": "10.1007/978-3-319-24574-4_28",
  "doiUrl": "https://doi.org/10.1007/978-3-319-24574-4_28",
  "requestedDoi": "10.1007/978-3-319-24574-4_28",
  "citationCount": 63195,
  "isOpenAccess": null,
  "openAccessUrl": null,
  "license": null,
  "abstract": null,
  "abstractNote": "No abstract available from this source.",
  "landingPageUrl": "https://doi.org/10.1007/978-3-319-24574-4_28",
  "bibtex": "@incollection{ronneberger2015unet_927a,\n  author = {Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas},\n  ...",
  "ris": "TY  - CHAP\nAU  - Ronneberger, Olaf\n...",
  "source": "Crossref",
  "attribution": "Metadata from the Crossref REST API (api.crossref.org), deposited by publishers. Abstracts, where included, remain under the work's licence shown in `license`."
}

The bibtex and ris strings are shortened above; in full they read:

@incollection{ronneberger2015unet_927a,
  author = {Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas},
  title = {{U-Net}: Convolutional Networks for Biomedical Image Segmentation},
  booktitle = {Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015},
  series = {Lecture Notes in Computer Science},
  year = {2015},
  pages = {234--241},
  publisher = {Springer International Publishing},
  address = {Cham},
  isbn = {9783319245737},
  doi = {10.1007/978-3-319-24574-4_28},
  url = {https://doi.org/10.1007/978-3-319-24574-4_28}
}

TY  - CHAP
AU  - Ronneberger, Olaf
AU  - Fischer, Philipp
AU  - Brox, Thomas
TI  - U-Net: Convolutional Networks for Biomedical Image Segmentation
T2  - Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015
T3  - Lecture Notes in Computer Science
PY  - 2015
SP  - 234
EP  - 241
PB  - Springer International Publishing
CY  - Cham
SN  - 9783319245737
DO  - 10.1007/978-3-319-24574-4_28
UR  - https://doi.org/10.1007/978-3-319-24574-4_28
ER  -

Notes on the fields:

• institutions holds affiliation text as publishers deposited it, often empty. authors lists up to 100 names, and authorCount gives the full number.
• venue is the first container title Crossref gives. For a chapter in a book series that is the series (as above); the BibTeX booktitle names the book itself.
• requestedDoi is the DOI you asked for, lower-cased (null in searches). doi is the DOI Crossref returned.
• isOpenAccess is true when the work carries a Creative Commons licence, and otherwise null (unknown, not "closed"): Crossref has no free-to-read flag.
• The dataset has two views: Works (an overview table) and Citations (BibTeX and RIS).

BibTeX and RIS

Both are built from the metadata already fetched: no extra requests, no extra cost.

Crossref type	BibTeX	RIS
journal-article	@article	JOUR
proceedings-article	@inproceedings	CPAPER
book, monograph, reference-book	@book	BOOK
edited-book	@book (editors, when it lists no authors)	EDBOOK
book-chapter, book-section, book-part, reference-entry	@incollection	CHAP
proceedings	@proceedings	CONF
report	@techreport	RPRT
dissertation	@phdthesis (@mastersthesis when the degree says so)	THES
posted-content (preprints)	@misc with a note	UNPB for preprints, else GEN
dataset and everything else	@misc with a note naming the type	DATA, STAND, ... or GEN

• Citation keys are the first author's surname, the year and the first significant title word, in plain ASCII, then _ and four characters derived from the DOI: ronneberger2015unet_927a. The same work gets the same key in every run, and two works never share one in practice.
• Names are "Last, First", joined by and. Corporate authors such as {The Gene Ontology Consortium} are braced so BibTeX keeps them whole. After 100 names BibTeX adds and others; RIS lists the first 100.
• Special characters are LaTeX-safe: & % $ # _ ~ ^ \ are escaped, and braces become \textbraceleft{} and \textbraceright{}, because BibTeX counts every brace, even an escaped one. Accented Latin letters become accent commands (Andr{\'e}, L{\ae}greid); other scripts stay UTF-8, which biber and a UTF-8 LaTeX handle. Title words with capitals after the first letter ({CCS}, {U-Net}, {mRNA}) are braced so styles keep their case; other proper nouns in titles aren't detected.
• DOI and link: doi holds the bare DOI, lower-cased and unescaped (styles print it verbatim), and url the doi.org link.
• Limits: search results don't carry Crossref's preprint subtype or a thesis's university, so a preprint found by search is labelled "Posted content", and a thesis names its publisher as the school. A DOI lookup gets both when Crossref has them.
• RIS uses one tag per line and ends each record with ER - . To build a .ris file, join the strings with a blank line between them.

Abstracts and licences

Abstracts are often the publisher's or the authors' copyright; Crossref warns that some "may be subject to copyright". So this Actor includes an abstract only when the work carries a licence that allows commercial reuse: CC BY, CC BY-SA, CC BY-ND or CC0/public domain. For every other work, abstract is null, and abstractNote explains why and points you to the DOI. When you reuse an included abstract, credit the authors and the licence: the record carries the authors, title, DOI and license you need. BibTeX and RIS strings don't include abstracts.

Data source, terms and limits

• Crossref (api.crossref.org): metadata that publishers deposit, which Crossref says you "may use […] for any purpose". The Actor follows Crossref's etiquette: it identifies itself and its operator's contact, sends one request at a time, follows the rate limit Crossref announces in each response (never faster than 5 a second; 3 a second in the polite pool on 2026-10-04), and backs off when asked. A search fetches 100 works per request. A DOI list takes one request per DOI, so 500 DOIs take a few minutes.
• Limits: at most 10,000 works per search and 500 DOIs per run. Results reflect Crossref's index on the day. Crossref's citation counts are usually lower than other indexes'.

The full terms, quoted with links, are in SOURCE-TERMS.md.

Pricing (suggested pay-per-event)

Event	Suggested price
Actor start (apify-actor-start)	$0.005
Result: one work saved to the dataset (apify-default-dataset-item)	$0.002 ($2 per 1,000 works)

Why these prices:

• They match the Store's common "1,000 results for $2", the same per-result price as Joi's SEC and TED Actors.
• Running costs are small: 1,000 works from a search take 10 API requests and well under a minute at 256 MB.
• DOIs not found and entries that aren't DOIs are never charged: they are not dataset items.

The run stops cleanly when it reaches the maximum cost you set for the run.

For the Actor's operator

Crossref asks API users for a contact address. Set it once in .actor/actor.json (environmentVariables.CONTACT_EMAIL, which ships as the placeholder [CONTACT EMAIL]), or in the Apify Console under the Actor's environment variables. While the placeholder is in place, every run stops with a clear message and makes no requests.

Local development

python -m unittest discover -s tests   # offline unit tests with recorded API responses, no network
CONTACT_EMAIL=you@example.com apify run --input '{"query":"kelp forest carbon","maxResults":10}'
CONTACT_EMAIL=you@example.com apify run --input '{"dois":["10.1039/c7ee02342a","doi:10.1109/CVPR.2016.90"]}'

The logic in src/core.py and src/citations.py uses only the Python standard library. src/main.py is a thin wrapper around the Apify SDK (apify==4.0.2).

Support

Report problems in this Actor's Issues tab. This is a search and lookup tool. The publisher's version of each work is the authoritative record.