How do I install the PDF Extract References skill?

Run npx skillmds@latest add jabref/pdf-extract-references in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is PDF Extract References safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with PDF Extract References?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is PDF Extract References free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.

Who published PDF Extract References?

jabref (@jabref) published this skill. Their other Agent Skills are listed on their SkillMD profile.

PDF Extract References by jabref — Agent Skill

Name: PDF Extract References
Author: jabref

PDF Extract References

Extracts the works a paper cites - parses the "References" section of PDF papers with JabRef's jabkit CLI and outputs the cited entries as BibTeX, via offline rule-based parsing (IEEE-style numbered references), a GROBID service, or an experimental LLM mode.

jabref b45adca 3.5 KB Updated Sep 17, 2026

File contents

Extract references from a PDF

Turn the "References" section of a PDF paper into BibTeX entries — one entry per cited work — using jabkit pdf extract-references.

This skill is about the works a paper cites. For a BibTeX entry describing the paper itself (its own title, authors, year), use the pdf-to-bibtex skill instead.

Setup

jabkit runs via JBang — no JDK setup needed. Install JBang if missing:

curl -Ls https://sh.jbang.dev | bash -s - app setup   # Linux/macOS

iex "& { $(iwr -useb https://ps.jbang.dev) } app setup"   # Windows

Then install jabkit on the PATH (the same command also updates it):

jbang app install --fresh --force jabkit@jabref

Basic usage

jabkit -p pdf extract-references paper.pdf                      # single PDF → BibTeX on stdout
jabkit pdf extract-references paper.pdf --output refs.bib       # single PDF → file
jabkit pdf extract-references a.pdf b.pdf --output-dir refs/    # one .bib per PDF into refs/
jabkit pdf extract-references a.pdf b.pdf                       # writes a.bib, b.bib next to each PDF
jabkit -p pdf extract-references https://example.org/paper.pdf  # each input may also be a URL

-p/--porcelain suppresses progress messages — use it whenever parsing the output. --output is only valid with a single input file; use --output-dir for several.

Inside --output-dir each .bib is named after its PDF. Inputs from different directories that share a file name do not overwrite each other: the second paper.pdf of a run is written as paper-2.bib, the third as paper-3.bib. Re-running the same batch replaces its own previous output instead of adding more numbered copies.

Every positional input may be a local path or an http(s)/ftp URL; URLs are downloaded to a temporary file first. For a URL input without --output/--output-dir, the default "next to the source PDF" target becomes the current working directory.

Extraction modes

Select with --mode (case-insensitive):

Mode	Strategy
`RULE_BASED`	Offline parsing of numbered `[1] ...` references (IEEE-style two-column papers). Default when no GROBID is configured.
`GROBID`	Sends the PDF to a GROBID service. Default if GROBID is enabled in preferences; `--grobid-url <url>` overrides the server for this call.
`LLM`	Experimental — uses the AI provider configured in preferences.

In RULE_BASED mode each entry keeps the original reference string in its comment field, so the parse can be double-checked against the source.

Notes

The exit code is non-zero if any input PDF was missing, could not be downloaded, or failed to parse; remaining files are still processed.
Reference parsing is heuristic. Entries with a doi field can be cross-checked by fetching clean metadata: jabkit doi-to-bibtex <doi>.

BibTeX entry for the paper itself: the pdf-to-bibtex skill (jabkit convert --input paper.pdf --input-format pdfMerged).
Cited works looked up online by DOI instead of PDF parsing: jabkit get-cited-works <doi>.
Full CLI reference: the jabkit skill in this repository.

---name: pdf-extract-referencesdescription: Extracts the works a paper cites - parses the "References" section of PDF papers with JabRef's jabkit CLI and outputs the cited entries as BibTeX, via offline rule-based parsing (IEEE-style numbered references), a GROBID service, or an experimental LLM mode.license: MIT---# Extract references from a PDFTurn the "References" section of a PDF paper into BibTeX entries — one entry per cited work — using `jabkit pdf extract-references`.This skill is about the works a paper *cites*. For a BibTeX entry describing the paper *itself* (its own title, authors, year), use the `pdf-to-bibtex` skill instead.## Setup`jabkit` runs via [JBang](https://www.jbang.dev/) — no JDK setup needed. Install JBang if missing:```bashcurl -Ls https://sh.jbang.dev | bash -s - app setup # Linux/macOS``````powershelliex "& { $(iwr -useb https://ps.jbang.dev) } app setup" # Windows```Then install `jabkit` on the PATH (the same command also updates it):```bashjbang app install --fresh --force jabkit@jabref```## Basic usage```bashjabkit -p pdf extract-references paper.pdf # single PDF → BibTeX on stdoutjabkit pdf extract-references paper.pdf --output refs.bib # single PDF → filejabkit pdf extract-references a.pdf b.pdf --output-dir refs/ # one .bib per PDF into refs/jabkit pdf extract-references a.pdf b.pdf # writes a.bib, b.bib next to each PDFjabkit -p pdf extract-references https://example.org/paper.pdf # each input may also be a URL````-p`/`--porcelain` suppresses progress messages — use it whenever parsing the output. `--output` is only valid with a single input file; use `--output-dir` for several.Inside `--output-dir` each `.bib` is named after its PDF. Inputs from different directories that share a file name do not overwrite each other: the second `paper.pdf` of a run is written as `paper-2.bib`, the third as `paper-3.bib`. Re-running the same batch replaces its own previous output instead of adding more numbered copies.Every positional input may be a local path or an `http(s)`/`ftp` URL; URLs are downloaded to a temporary file first. For a URL input without `--output`/`--output-dir`, the default "next to the source PDF" target becomes the current working directory.## Extraction modesSelect with `--mode` (case-insensitive):| Mode | Strategy ||---|---|| `RULE_BASED` | Offline parsing of numbered `[1] ...` references (IEEE-style two-column papers). Default when no GROBID is configured. || `GROBID` | Sends the PDF to a [GROBID](https://github.com/kermitt2/grobid) service. Default if GROBID is enabled in preferences; `--grobid-url <url>` overrides the server for this call. || `LLM` | Experimental — uses the AI provider configured in preferences. |In `RULE_BASED` mode each entry keeps the original reference string in its `comment` field, so the parse can be double-checked against the source.## Notes- The exit code is non-zero if any input PDF was missing, could not be downloaded, or failed to parse; remaining files are still processed.- Reference parsing is heuristic. Entries with a `doi` field can be cross-checked by fetching clean metadata: `jabkit doi-to-bibtex <doi>`.## Related- BibTeX entry for the paper itself: the `pdf-to-bibtex` skill (`jabkit convert --input paper.pdf --input-format pdfMerged`).- Cited works looked up online by DOI instead of PDF parsing: `jabkit get-cited-works <doi>`.- Full CLI reference: the `jabkit` skill in this repository.

Frequently asked questions