English | 简体中文
Give ScholarFetch a research topic, a reference list, or a DOI; it returns verified records and the publishers' own PDFs. The Skill and its command-line tool work in three stages:
- Search Crossref, PubMed, OpenAlex, Europe PMC, Semantic Scholar, and arXiv, merging duplicates into one candidate list with source records.
- Verify titles, authors, years, and DOIs against the supplied references, flagging conflicts and missing fields.
- Download PDFs from official publishers through a dedicated browser that navigates each publisher's page with your own institutional access. Each PDF is identity-checked before it is accepted.
Every result and every failure is recorded.
Note
Human-in-the-Loop. This Skill handles repetitive research tasks; it supports research rather than doing it. Keep questions, judgments, and conclusions with people, and check every result before use.
A topic runs the full chain, a reference list starts at verification, and a DOI or publisher link goes directly to download. Verified records can be exported as RIS or BibTeX.
Search results are candidates; which papers to cite remains the user's decision. Inserting citations into Word is handled by Zotero-Word-AutoCite.
Install from the repository root and check the environment:
python -m pip install .
literature-workflow doctor --mode all- The base install (Python 3.11 or later) covers search, verification, direct downloads, and PDF checks.
- Browser downloads, the main acquisition route, need
python -m pip install ".[browser]"and the cloakbrowser binary. Without them, the other stages still run. - Some sources need configuration, such as
PUBMED_EMAILfor PubMed;doctorreports what is missing. - Restricted papers use your existing publisher access.
No MCP server, external language model, or paid relay is required. Accounts, cookies, browser profiles, and browser binaries are not bundled. See setup and dependencies.
Place the repository in your agent's skill directory as scholarfetch (for Codex, usually ~/.codex/skills/scholarfetch/), then ask:
Use $scholarfetch to find recent research on excitons in two-dimensional materials.
Report the sources and search scope, verify the candidates, and download their official PDFs.
You can set dates, result counts, preprint preferences, and the output directory. Inputs can be text, JSON, CSV, or DOI-bearing RIS/BibTeX; see request examples. The same stages are available from the command line:
literature-workflow search "exciton transport two dimensional materials" --sources crossref,europe_pmc --rows 3 --language en --output runs/search
literature-workflow verify --doi "10.1371/journal.pone.0000308" --language en --output runs/verify
literature-workflow download --input papers.txt --no-browser --language en --output runs/download
literature-workflow search-download "exciton transport two dimensional materials" --rows 3 --no-browser --language en --output runs/search-download
literature-workflow export --input runs/search-download/records.json --format ris --language en --output runs/exportUse a new or empty output directory for each run; existing results are never overwritten. To check an original reference rather than only resolve a DOI, add --expected-title and --expected-year. See --help for all options.
Each run writes run.json, records.json, and report.md. Downloads add download_manifest.json and put accepted PDFs in papers/; anything unavailable or uncertain stays in the records and report with a reason. Exports write references.ris or references.bib. Fields and statuses are defined in the workflow contract.
Searches are bounded and do not promise exhaustive coverage. Automated acquisition accepts verifiable official publisher sources only, with no fallback to repository or third-party copies. Each browser session uses a fresh isolated profile, and credentials stay local.
65 offline tests pass, covering source errors, identifier and version conflicts, PDF identity checks, overwrite protection, and sensitive-information handling. Live checks with Crossref and Europe PMC completed a search → verify → download → RIS export run that produced an accepted official PDF. Conflicting records were kept out of exports, and signed URL parameters and email addresses were removed from RIS/BibTeX.
These checks do not establish support for every source or publisher.
ScholarFetch integrates discovery and verification code from nature-academic-search, source-operation knowledge from paper-lookup in K-Dense Scientific Agent Skills, and an official PDF acquisition workflow. This project adds a unified command interface, stage handoffs, dependency diagnostics, and traceable outputs. Upstream knowledge coverage does not imply that every upstream database is implemented.
New project code is released under MIT. Upstream code and runtime dependencies retain their own licenses. See third-party notices for contributions and preserved license texts.