How Paperful compares

A plain-language map of where Paperful sits next to Zotero plugins, bibliography fixers, Mendeley export cleaners, DOI-centric download scripts, and the 2026 cluster of hosted literature / AI-scientist / manuscript tools.

1.x is still a disk-first Zotero helper whose public job is find missing PDFs. Hosted “paper factories” (Edison, Elicit, Deep Research, Jenni) own discover and draft; they do not own campus-proxy fetch, grey playbooks, or a restore ledger. Factory-stage matrix: comparison reference.

Last reviewed: 2026-10-27. Feature lists for other products are based on public docs and positioning — not paid pilots or exhaustive release testing.

Vendor-by-vendor notes live in the comparison reference. Prefer this page for “is this the right tool?”

Short answer

If you need…	Look at…
A platform-agnostic mirror of the library (backup, and a way out if the citation manager changes). Zotero is well tested. Mendeley and EndNote adapters are seeking testers	Paperful snapshot / restore — quiet mirror. Do not treat the other adapters as proven
Bulk missing-PDF fetch for a Zotero library (open access first, campus proxy when you have it), collection-shaped folders, resumable CLI	Paperful (this repo)
In-Zotero “find OA PDF” plus optional grey-zone sources in one plugin UI	zotero-zotadata
Attachment hygiene (broken links, rename, stored-to-linked, duplicate files on one parent)	Paperful attachments (report by default; surgery with a flag and --apply). In-app: StorScan, Attanger, ZotMoov
Metadata repair (DOI/ISBN/arXiv bulk update, parent-from-PDF)	Paperful lint / fix-metadata, or ZotMeta
Duplicate parents in one collection (DOI, then title+year). Review a pack, then merge the PDF, notes, and better fields onto one item.	Paperful dedupe
Grey literature landings (UN, FAO, ISA, and similar) kept as real PDFs	Paperful playbooks in direct / landing. Journal-style OA fetch is the row above
Batch notes from PDFs you already have: one grounded summary per item, then a collection review. Local model, off by default. Scans need ocr first	Paperful summarize / synthesize
Scriptable library surgery (merge, enrich, disk GC, two-up scan split) via CLI/MCP	zotero-agent
AI assistant read/write over the library, including chat and (on some forks) OCR of scans	zotero-mcp forks (richardjlyon, cookjohn, mcp-zotero)
Semantic search, citation graph, author/trend tools over the public literature via MCP (hosted corpus; not your Zotero tree)	Valency MCP Bond — see reference
Hosted literature search / cited review over a public corpus (not your Zotero tree)	Elicit, Consensus, Undermind, SciSpace, Scite — see reference
Hosted AI scientist — cited reviews over ~175M papers/trials/patents, “has anyone done this?”, optional multi-hour discovery + experimental analysis	Edison Scientific (Literature / Precedent / Kosmos). Nearby loops: Google Co-Scientist, Sakana AI Scientist (workshop-grade, not a library helper) — reference
Web Deep Research memo (cited longform from the open web in minutes; not a journal manuscript)	OpenAI / Gemini / Perplexity Deep Research. Paperful Ask is over held PDFs
Systematic-review screening (dual review, extraction forms, PRISMA counts; usually no discovery corpus)	Covidence, Rayyan, DistillerSR; local ASReview. Paperful sr can export counts and CSV; it does not replace those tools. Discover Keep/Skip is hop triage, not dual review
Chat with PDFs you upload (no scholarly search, no Zotero restore)	Scholarcy, Explainpaper, Humata, Gemini Notebook. Paperful ask uses the same out/ tree as fetch
Draft a manuscript with citations (library ∪ global index; language/LaTeX)	Jenni, Paperpal, SciSpace Writer, Writefull in Overleaf. Living cites without a model: Better BibTeX → Quarto. Paperful does not draft papers. rewrite applies an existing critique pack as locatable edits / tracked changes — it is not a generator
Apply a pre-submission pack as tracked changes (hygiene first; no invented numbers or cites)	Paperful rewrite from a critique pack. Patterns (not vendored code): Paper Orchestra worklog, ARIES align-then-generate, XtraGPT selected-span prompts
Grounded Q&A over PDFs you already have (local index; same collection mirror as fetch)	Paperful rag ingest / ask (simple default; --ask-mode advanced scores passages then answers) — optional paperful mcp (ask, refs_gap, gaps, snowball_search, snowball_trends, export, proposal_export). OSS cousin: PaperQA2 (you supply the PDFs)
Deep forensics on tables, stats, and figures (GRIM/statcheck/TIVA, spreadsheet digit tests, image duplication, paper-mill graphs) — flags anomalies with innocent explanations, not verdicts	PaperGuard (paperguard scan). Paperful critique --forensics wraps that CLI when it is on PATH. screen / critique cover retractions, PPS fingerprints, and light stdlib stats on manuscript text — complement, not a substitute. Later buffet items: Research integrity forensics
.bib normalize / dedupe / upgrade preprints (no Zotero required)	bibcite, bibtex-tidy, bibmanager
Mendeley dedup inside the app; clean exported BibTeX	Mendeley Duplicates smart collection; export cleaners such as mendeley_bibtex_cleaner
DOI-list PDF batch without Zotero	paperscraper
Grow the library from a keyword, one or more DOIs’ references, one or more ORCIDs, a similar-paper hop, or a hybrid hop, then optionally fill PDFs. Citation maps such as Research Rabbit link out; Paperful downloads. Re-check later with watch (baseline once, then propose new arrivals on disk; you schedule watch run)	Paperful snowball — dry-run, approve-each, approve-batch, --gate auto, --fetch-pdfs, and watch (snowball). In-app one-hop browsers stay separate (zotero-snowball, Citegeist). General harvesters without the mirror: findpapers, opencite
People you follow → their new papers (ORCID / OpenAlex; no hop; corpus suggest; briefing with recent works / co-authors; saved social HTML/CSV, no live scrape)	Paperful authorwatch (authorwatch). Distinct from snowball watch.
Alert loop / weekly what-cites (poll watches + citers of -C on a cadence; apply into the library; no email; you cron it)	Paperful keepup (keepup)
Local workbench over the same CLI (Wanted / Discover / Ask; Advanced Repair, Mirror, Index, Briefs)	Paperful docker compose up or paperful serve — gui.md. Grab fetches to disk; Attach writes the library. Zotero stays the catalogue. Browser login stays the CLI
Save the page open in Firefox into the folder copy. Keep the Zotero Connector for saving into Zotero	Paperful Firefox add-on, optional and unsigned. Writes out/ only. No PDF on the page: HTML snapshot, plus your configured search lanes if you ask. Several selected tabs at once, reviewed before save. Context menu and Alt+Shift+P
“Just use what ships in Zotero”	Built-in Find Available PDF plus custom PDF resolvers

Paperful’s jobs are library, find, completeness, mirror, and control. Fetch and lint run on disk; the manager is a write-back adapter (manager = "zotero" is well tested; Mendeley and EndNote are seeking testers). attachments reports layout problems and, with a flag plus --apply, repairs them from out/. --link is that repair. For a full metadata editor, phone sync, mobile apps, WebDAV, incoming downloads, author folders, or tablet send/get, use the citation manager or the tools named in the table (Attanger, ZotMoov, and the .bib linters).

Architecture: architecture.md.

Where Paperful sits

Most tools in this space optimise one or more of:

1. Acquire PDFs — open access, proxy, optional Sci-Hub, Scholar

2. Fix metadata — DOI discovery, Crossref/OpenAlex fills

3. Fix files — rename, linked paths, broken attachments, duplicate PDFs

4. Fix .bib / exports — keys, duplicates, Mendeley-specific fields

5. Automate / agent — MCP, CLI, batch undo

Those map onto Paperful’s jobs. Find is (1): open access, campus proxy, playbooks, and an opt-in AI browser after the scripted lanes fail. Completeness is (2) (lint / fix-metadata, patches on disk, --apply to the adapter) plus duplicate review (dedupe) and opt-in summarize / synthesize. Mirror is the portable disk copy (snapshot / restore; snapshot --pdfs all also copies PDFs already in Zotero). (3) is paperful attachments: a report by default, and --fix-broken, --merge-files, --rename, or --link only with --apply. (4) is other tools. (5) here is CLI --format json (paperful.agent.json.v1 + exits) on batch verbs, plus a thin optional paperful mcp for read-only agent tools (gaps, snowball keyword dry-run, scoped export, proposal-pack BibTeX/RIS, refs_gap, ask, precedent) — same JSON as --format json. A chat agent that mutates the live Zotero catalogue stays with zotero-mcp. Scanned PDFs get a text layer from paperful ocr; two-up split and shrink stay with zotero-agent pdf-prep. Library is the catalogue you already have, grown with snowball when you ask. Control is disk-first write-back and opt-in sources.

Reference library  →  paperful snapshot  →  out/<collection>/<stem -- KEY>/
                 →  paperful run       →  same folder (PDF)  →  attach (Zotero 10+)
                 →  paperful restore --apply  →  missing items only
                      ↑
        OA / CORE / arXiv / EZProxy / Scholar / htmlpdf / grey playbooks / (opt-in Sci-Hub)

Optional, off until [llm].enabled (needs a text layer; `paperful ocr` adds one):
  summarize · synthesize · recover (browser agent; last run lane after vault browsers fail)

Parallel tracks:
  Metadata: paperful lint/fix-metadata, ZotMeta, zotero-agent
  Duplicates: paperful dedupe (merge onto the keeper, then trash the extra), Zotero’s duplicate UI, zotero-agent
  Attachments: paperful attachments (CLI). In-app: StorScan, Attanger, ZotMoov
  Chat: zotero-mcp. Two-up scan split: zotero-agent pdf-prep
  Text layer for image PDFs: paperful ocr
  Bib CLI (bibcite, bibtex-tidy)

Capability snapshot

Legend: Yes = first-class · Partial = adjacent or lighter · No = absent or out of scope.

Capability	Paperful	Zotero built-in	zotero-zotadata	StorScan	ZotMeta	zotero-agent
Bulk fetch missing PDFs	Yes	Partial	Yes	Partial	No	Partial
Collection-scoped batch runs	Yes	No	Partial	Partial	Partial	Yes
Resumable manifest / retry	Yes	No	Partial	Partial	Partial	Partial
Open-access source stack	Yes	Partial	Yes	Partial	No	Partial
Campus EZProxy (cookie session)	Yes	No	No	No	No	No
Sci-Hub (explicit opt-in)	Yes	Partial	Yes	No	No	Partial
Metadata verify / lint	Yes	No	Yes	No	Yes	Yes
Metadata apply to library	Yes (fix-metadata --apply)	No	Yes	No	Yes	Yes
Broken link / file layout repair	Partial (attachments; report by default, surgery with --apply)	Partial	Partial	Yes	No	Partial
Dedupe / merge items	Yes (dedupe --apply merges children and better fields, then trashes the extra; Zotero only)	Partial	No	Partial	No	Yes
Runs outside Zotero UI (CLI)	Yes	No	No	No	No	Yes
Work on disk, then write-back	Yes	No	No	Partial	No	Partial
Per-item folder you can restore from	Yes (snapshot / restore; does not overwrite fields)	No	No	No	No	No
Copy PDFs already in Zotero onto disk	Partial (snapshot --pdfs all)	Partial (File → Export PDFs)	No	No	No	No
Grey-literature landing playbooks	Yes	No	No	No	No	No
Grounded summary / collection review	Partial (opt-in, local, text layer)	No	No	No	No	Partial (public README: summarize PDFs into notes)
OCR for scanned PDFs	Partial (ocr, OCRmyPDF text layer on disk)	No	No	No	No	Partial (pdf-prep, OCRmyPDF)

zotero-mcp, Valency, Edison, Elicit/Consensus, SR tools, writers, and BibTeX columns: comparison reference. Factory stages (discover → PDF → screen → write): paper factory stages.

What Paperful does not do today

• Mendeley and EndNote adapters exist and are seeking testers. Zotero is the well-tested path. EndNote writes are an import bundle (File → Import); Paperful does not edit the .enl database

• Author-folder layouts, incoming-download matching, and tablet send/get (attachments --link can point a personal library at out/; it is off unless you pass it)

• Mendeley and EndNote item merge (dedupe --apply is Zotero-only)

• Two-up scan split, or a chat agent over the library (ocr adds a text layer; it does not split pages. summarize / synthesize / recover are opt-in and local)

• Hosted multi-user service

• Autonomous AI-scientist runs, experimental-data kernels, or molecular design (use Edison Scientific or similar)

• A journal manuscript, related-work draft, or “overnight PhD” (use Jenni/Paperpal/Overleaf for drafting; Edison/Sakana are not Paperful’s job)

• Systematic-review dual screening and PRISMA-grade recall (use Covidence / Rayyan / ASReview; Discover Keep/Skip is hop triage)

• Jeffersonian transcription or qualitative coding

Choosing in one glance

Have Zotero, many items without PDFs, want a resumable CLI + folder tree?
  → paperful

Want the files and records on disk, independent of Zotero cloud quota?
  → paperful snapshot (restore recreates only missing items)

Want grounded notes, then one review of a collection, without a chat session?
  → paperful summarize, then synthesize (local model; text-layer PDFs)

Library DOI looks wrong but you already have the PDF?
  → paperful lint, then fix-metadata --apply

Want the same job but stay inside Zotero with one plugin?
  → zotero-zotadata (review legal/ToS for its extra sources)

Library is messy on disk (broken links, dup PDFs on one item, filenames)?
  → paperful attachments (report). Add --fix-broken, --merge-files, --rename, or --link with --apply to write.
  Author folders, incoming downloads, tablet send/get → StorScan, Attanger, ZotMoov

Only a .bib from Mendeley/Zotero export?
  → bibcite / bibtex-tidy / JabRef

Want a new collection from a keyword, one or more papers’ bibliographies, one or more ORCIDs, or one hop from the top hits?
  → paperful snowball (see snowball.md). Pass several DOIs or ORCID iDs in one command. One-shot PDF fill is --fetch-pdfs with --gate auto

Want new works matching a saved snowball profile, without a discovery daemon?
  → paperful snowball watch (baseline once, then propose). The rollup is `snowball watch digest` or `watch run --digest`. Schedule that yourself ([Watch](snowball.md#watch))

Want new papers from people you already follow (ORCID / OpenAlex ids, no hop)?
  → paperful authorwatch (see authorwatch.md). `suggest -C` ranks corpus people; import CSV or a saved RG/LinkedIn/Academia HTML page (`--file`). No live social scrape.

Just a list of DOIs, no Zotero?
  → paperscraper

Scanned PDFs with no text layer?
  → paperful ocr (disk text layer). Two-up books: zotero-agent pdf-prep

Want a chat agent in the loop?
  → a zotero-mcp fork

Need semantic search / citation graph / trends across all public papers, not just your library?
  → Valency MCP Bond (or similar lit MCP). Paperful snowball + authorwatch grow *your* collection; `snowball trends` for year histograms on a query; `mcp` / CLI JSON for gaps, dry-run discovery, trends, export, and ask over *your* mirror

Need a hosted cited review, “has this been tried?”, or a Kosmos-style run over a dataset plus the public literature?
  → Edison Scientific (Literature / Precedent / Kosmos). Paperful still owns PDFs, the disk mirror, and Ask over *your* collection

Need Elicit/Consensus-style tables over the public record, not campus PDFs?
  → Elicit, Consensus, Undermind, SciSpace. Then snowball / ingest-dois → paperful run

Need dual-review screening and a PRISMA flow?
  → Covidence, Rayyan, DistillerSR, or local ASReview. Paperful fills PDFs and can later export counts

Need a Deep Research memo from the web?
  → OpenAI / Gemini / Perplexity. Not a Zotero collection and not a journal paper

Need a manuscript draft with extra cites from a huge index?
  → Jenni, Paperpal, SciSpace Writer, Overleaf AI. Paperful will not invent BibTeX; extra cites belong in Zotero first. `paperful rewrite` only applies a critique pack you already ran

Need locatable edits from a critique pack (hygiene / tracked-changes Word)?
  → paperful rewrite (see rewrite.md). Not a generated paper

Need cited Q&A on a folder of PDFs with no Zotero?
  → PaperQA2 (OSS). Paperful when those PDFs should also live in a restore ledger

More branches: comparison reference. How far a snowball hop reaches: How a hop is cut.

Related docs

• architecture.md — disk-first adapters, circuit breaker, Sci-Hub

• ROADMAP.md — 1.x Wanted + mirror; v2 factory packs after the ledger is honest

• snowball.md — keyword, multi-DOI / multi-ORCID, collection, and hybrid library building

• authorwatch.md — people lists → new papers (suggest / accept / import; no hop)

• comparison-reference.md — vendor notes and extra tables

• commands.md — CLI and disk artifacts

• config.md — config.toml and operations