PaperHarbor

Finding full texts, naming files, and tracking missing papers interrupts literature review. A bibliography may contain a mixture of publisher links, repository copies, and references with incomplete metadata, making manual collection repetitive.

Import Zotero, BibTeX, CSL-JSON, or a DOI/URL. Search and select references locally, then download available copies. The same discovery pipeline supports interactive selection and repeatable command-line batches.

The provider aggregator checks reviewed author copies and supported publisher links, then conditionally queries Unpaywall, OpenAlex, Europe PMC, arXiv, Semantic Scholar, and Crossref. If initial downloads fail, it queries providers skipped earlier, including search fallbacks, and inspects publisher pages.

Candidate URLs are deduplicated, direct PDFs receive priority over landing pages, and conflicting DOI candidates are rejected. Provider errors are recorded while other searches continue.

The application coordinates readers, provider searches, landing-page resolution, file naming, downloads, and reporting. It checks existing attachments and destination files before searching. Repeated DOI outcomes within a batch are reused to avoid duplicate network requests.

References are processed sequentially with a default 0.5-second pause. An individual reference error is recorded and processing continues with the next paper.

• Require a PDF signature within the first 1,024 bytes and a minimum download size of 1,024 bytes.
• Reject redirects that identify a conflicting DOI.
• For candidates requiring an identity check, verify the requested title against text from the PDF’s first page.
• Stream to a temporary file before moving a successful download into its destination.

Transient network failures and selected HTTP responses receive up to two retries by default. Backoff uses the server’s Retry-After value when available; requested waits above 30 seconds are deferred for a later attempt.

Access-denied responses are not repeatedly retried. A PDF header check identifies the file format; it does not establish the scientific quality or completeness of a paper.

• Downloaded or already present in Zotero or the destination.
• No open copy found, download failed, or manual access required.
• Reference error, recognized web resource, or withdrawn paper.

Pass a previous report with --retry-report to revisit failed, manual-access, and error rows. Add --retry-not-found to include papers for which no copy was found. Verified alternate URLs can be supplied through --pdf-overrides.

.\.venv\Scripts\python.exe main.py `
  --export-file "references.bib" `
  --dest-folder "D:\Research\Papers" `
  --limit 10

.\.venv\Scripts\python.exe main.py `
  --retry-report "previous_report.csv" `
  --dest-folder "D:\Research\Papers"

The server binds to localhost. The SQLite reader locates common Zotero database paths or accepts a custom path. It queries a temporary copy and attempts to copy accompanying WAL and SHM files when present, then closes the connection and removes the snapshot.

Downloads stay in your chosen folder. PaperHarbor does not attach them to Zotero or update library records. Reference metadata is sent to discovery services during searches.

Access restrictions or missing metadata can prevent retrieval. Unresolved items retain follow-up links for manual review. Missing attachment detection relies on Zotero metadata, which may differ from files actually present on disk.

Sequential processing favors a traceable workflow over parallel throughput. An unsuccessful search does not establish that a paper is paywalled. Publisher access rules still apply.