How Paperful compares — reference

Vendor-by-vendor notes and longer capability tables. Most readers only need How Paperful compares.

Last reviewed: 2026-10-26. Feature lists for other products are based on public docs and positioning — not paid pilots.

Capability snapshot

Zotero-centric (extended)

Legend: Yes · Partial · No.

Capability

Paperful

Zotero built-in

zotero-zotadata

StorScan

ZotMeta

zotero-agent

zotero-mcp (typical)

Bulk fetch missing PDFs

Yes

Partial

Yes

Partial

No

Partial (pdf-fetch)

Partial (Unpaywall attach)

Collection-scoped batch runs

Yes

No

Partial

Partial

Partial

Yes

Partial

Resumable manifest / retry

Yes

No

Partial

Partial

Partial

Partial

No

Open-access source stack

Yes

Partial

Yes

Partial

No

Partial

Partial

Campus EZProxy

Yes

No

No

No

No

No

No

Sci-Hub (explicit opt-in)

Yes

Partial

Yes

No

No

Partial

No

Metadata lint / verify DOI

Yes

No

Yes

No

Yes

Yes

Partial

Metadata apply

Yes

No

Yes

No

Yes

Yes

Partial

Attachment / disk repair

No

Partial

Partial

Yes

No

Partial

No

Dedupe / merge items

Yes (dedupe --apply merges children and better fields, then trashes the extra; Zotero only)

Partial

No

Partial

No

Yes

Partial

CLI / automation

Yes

No

No

No

No

Yes

Yes (MCP)

Disk-first then write-back

Yes

No

No

Partial

No

Partial

No

Per-item folder you can restore from

Yes

No

No

No

No

No

No

Copy PDFs already in Zotero onto disk

Partial (snapshot --pdfs all)

Partial (File → Export PDFs)

No

No

No

No (export is bibliography; get_item_pdf_path is a path)

Partial (get_pdf_path)

Grey-literature landing playbooks

Yes

No

No

No

No

No

No

Grounded summary / collection review

Partial (opt-in summarize / synthesize; text layer)

No

No

No

No

Partial (README: summarize PDFs into notes)

Yes (chat reads PDF text; the model writes the summary)

OCR for scanned PDFs

Partial (paperful ocr, OCRmyPDF text layer on disk)

No

No

No

No

Partial (pdf-prep, OCRmyPDF)

Partial (some forks, e.g. Docling OCR pre-step)

Zotero version and attach (indicative)

Verify against each project’s latest release before upgrading Zotero.

Tool

Typical Zotero

Notes

Paperful

7–9 download; 10+ attach and fix-metadata --apply

Local HTTP API

StorScan, Attanger, ZotMeta 2.0, zotodata

7–9 (per upstream READMEs)

Plugin .xpi

zotero-agent

7+

Bridge plugin + CLI

ZotFile

≤6

Not maintained for Zotero 7+

Built-in Find Available PDF

6+

Custom resolvers in Config Editor

BibTeX / export fixers

Capability

bibcite

bibtex-tidy

bibmanager

bibtex-cleaning-agent

JabRef

Dedupe entries

Yes

Yes

Yes

Yes

Yes

Resolve DOI/arXiv/title → BibTeX

Yes

No

Partial (ADS)

Yes (LLM + APIs)

Partial

Stable citation keys

Yes

Partial

Yes

Yes (patterns)

Yes

Upgrade arXiv → published

Yes

No

Yes

Partial

Partial

Mendeley export cleanup

Partial

Partial

Partial

Partial

Yes (import)

Zotero library sync

No

No

No

No

Yes

Batch PDF download

No

No

Partial (ADS)

No

Partial

DOI-centric downloaders

Capability

Paperful

paperscraper

pyzotero + custom scripts

Needs a library manager

Adapter (Zotero now)

No

Optional

Input

Library items / collections

DOI JSONL / dict

API keys or local DB

Publisher TDM APIs

No

Yes (Elsevier/Wiley, etc.)

DIY

Attach back to items

Yes

No

DIY

EZProxy / Sci-Hub

Yes / opt-in

No / DIY

DIY

Hosted literature MCP (global corpus)

Legend: Yes · Partial · No. Valency rows reflect public positioning for MCP Bond (not a hands-on product audit). Edison rows reflect public docs (2026-10-26), not a paid pilot.

Capability

Paperful

Valency MCP Bond

Edison Scientific

zotero-mcp (typical)

Semantic search over public papers / preprints

Partial (snowball keyword / similar hops; OpenAlex-backed discovery — not a single “ask the whole web” MCP)

Yes

Yes (Literature / PaperQA3 over 175M+ papers; also trials and patents)

Partial (search library + sometimes Unpaywall)

Semantic search over your PDFs

Yes (rag ingest + ask; GUI Ask)

No

Partial (upload / workspace files; not your Zotero tree)

Yes (chat reads attached PDFs)

Citation graph (corpus-scale, both directions)

Partial (snowball refs / cites / both; Discover map of one queue, not the whole public corpus)

Yes

Yes (Literature: native citation-graph understanding)

Partial

Author profiles / co-author networks

Partial (paperful network ORCID ego + Discover People map; authorwatch briefing, suggest -C)

Yes

Partial (via literature search, not a dedicated author MCP)

Partial

Keyword / publication trends

Partial (snowball trends, MCP snowball_trends; OpenAlex filters in snowball)

Yes

Partial (reviews, not a trends MCP)

No

Novelty / “has anyone done this?”

Partial (refs gap, similar hops, ask --focus gaps — never claims absence)

No

Yes (Precedent, formerly Has Anyone)

No

Cited literature synthesis / conflicting evidence

Partial (opt-in synthesize / Ask over held PDFs)

Partial (search/export, not a review agent)

Yes (Literature; LITERATURE_HIGH reasoning mode)

Partial (chat)

Bulk missing-PDF fetch into a collection

Yes

No

No

Partial

Disk mirror + restore ledger

Yes

No

No

No

Campus EZProxy / grey-lit PDF playbooks

Yes

No

No

No

Experimental-data analysis / molecular design

No

No

Yes (Analysis, Molecules; Kosmos joins literature + code)

No

Programmatic API

Partial (paperful mcp + --format json; writes stay CLI --apply)

Yes (MCP)

Yes (edison-client jobs; Kosmos not on the public API)

Yes (MCP)

MCP tools (indicative breadth)

Partial (read-only: gaps, snowball_search, snowball_trends, export, proposal_export, refs_gap, ask, precedent; no --apply)

Yes (positioning: large tool surface)

No (Python/REST client; chat sessions)

Yes (varies by fork)

Runs local / library stays on disk

Yes

No

No (hosted platform; academic free tier)

Yes

Corpus freshness (vendor SLA)

Partial (snowball watch, authorwatch run — you schedule polls)

Yes (positioning: index within hours)

Yes (hosted index; SLA not audited here)

N/A (local library)

Valency Hub (publish, fork, public review) is a separate product; Paperful does not host papers. Edison Kosmos is a separate autonomous-scientist job, not an MCP Bond analogue. Docs say Literature covers 175M+ works; Edison’s February 2026 PaperQA3 post says 150M+ with figures — do not reconcile those numbers in a customer table.

Paper-factory stages (Oct 2026)

Nobody ships discover → campus PDF → dual screen → extract → synthesize → novelty → journal manuscript → keep current, and local/offline Zotero restore. Hosted tools saturate discover/synthesize. Paperful 1.x owns get PDF (proxy + grey + provenance) and the restore ledger. Scores are public-docs positioning (2026-10-26), not a paid audit.

Legend: Yes · Partial · No. Novelty is split from synthesize on purpose (Edison Literature vs Precedent). Write MS means a submission-shaped draft, not a Deep Research memo. Independent 2025 work found Elicit Pro Review-mode sensitivity ~40% vs traditional searches ~95% — do not mark Elicit screen as Yes.

Platform

Discover

Get PDF

Screen

Synthesize

Novelty

Write MS

Keep current

Local

Zotero

Paperful 1.x

Partial (snowball / authorwatch)

Yes (OA, CORE, EZProxy, grey; Sci-Hub opt-in)

Partial (screen; Discover Keep/Skip ≠ dual review)

Partial (summarize / synthesize / ask on held PDFs)

Partial (refs gap; never claims absence)

Partial (stuff CLI pack; no auto-insert)

Yes (you cron keepup)

Yes

Yes

Edison Literature

Yes (hosted PaperQA3)

Partial (OA/licensed corpus; no campus fetch)

No

Yes

No (that is Precedent)

Partial (cited report)

Partial (hosted index)

No

No

Edison Precedent

Yes

Partial

No

No

Yes (not a legal prior-art search)

No

Partial

No

No

Edison Kosmos

Partial (inside the run)

No

No

Yes (reports; paper: interpretation weaker than quotes)

Partial

Partial (report, not a Zotero MS)

No

No

No

Elicit

Yes (~138M)

Partial (upload / OA)

Partial (SR UI; low independent recall)

Yes

No

Partial (report templates)

Partial

No

Yes (import)

Consensus

Yes

Partial (OA + uploads; LibKey = links)

Partial

Yes

No

No (will not write your literature review)

Partial

No

Yes (one-way)

Undermind

Yes

Partial

No

Yes

No

No

Yes (alerts)

No

No

SciSpace

Yes

Partial

Partial (claimed SLR)

Yes

No

Yes (AI Writer)

No

No

Partial

Scite

Yes

No

No

Partial

Partial (supporting/contrasting)

No

No

No

No

Valency Bond

Yes

Partial (Hub = OA)

No

No

No

No

Yes

No

No

Covidence / Rayyan

No

Partial (imported PDFs)

Yes

Partial / No

No

No

No

No / Partial

Yes / Partial

ASReview

No

No

Yes (local active learning)

No

No

No

No

Yes

Partial

Jenni / Paperpal

Partial (200–300M + uploads)

Partial

No

Partial

No

Yes (drafting workspace)

No

No

Jenni: one-way

OpenAI / Gemini / Perplexity DR

Yes

Partial (web / uploads)

No

Yes (report)

No

No

No

No

No

Sakana AI Scientist

Partial

No

No

Partial

Partial

Yes as workshop-grade LaTeX

No

Partial (you run it; LLM APIs)

No

PaperQA2 (OSS)

No (you supply PDFs)

Partial

No

Yes (cited Q&A)

No

No

No

Yes

Partial (paper-qa[zotero]; last test is old)

Gemini Notebook / Humata

No

No

No

Yes (on uploads)

No

Partial

No

No

No

Sources and caveats: repo reports/AI academic paper factories.md (contributor research, Oct 2026).

PaperGuard (research integrity screener)

  • Sites: GitHub · PyPI · live demo

  • Fit: Offline-first anomaly screener for tabular data, reported statistics, images, and manuscript metadata. ~41 built-in detectors (GRIM/GRIMMER/SPRITE, statcheck, TIVA/P-curve, Carlisle RCT imbalance, tortured phrases, stylometry, opt-in AI-text classifiers, paper-mill citation-graph subgraph, image pHash / splice forensics, EXIF/rsid checks, Retraction Watch / PubPeer / ORI cross-checks). Batch CLI, HTML/JSON reports, optional Web UI and LLM explain. MIT license.

  • Epistemic line: flags statistical anomalies, mandatory innocent explanations, no “fraud” wording in reports — aligned with Paperful screen / critique copy.

  • With Paperful: Paperful owns Zotero mirror, fetch, and light manuscript packs; PaperGuard owns deep forensics on files (especially supplementary CSV/XLSX and raster figures). Overlap is narrow: retraction/PubPeer/tortured phrases — Paperful screen + critique pps_* on PDF text; PaperGuard adds full statcheck, table digit tests, and image duplication. paperful critique --forensics subprocesses paperguard when it is on PATH. Collection batch and workbench HTML deep-links remain roadmap.

  • Not a substitute for: Missing-PDF fetch, snowball discovery, Ask/RAG over your library, or institutional integrity investigations.

Capability

Paperful

PaperGuard

Retraction / EoC on library items

Yes (screen, critique cited works)

Yes (DOI cross-check)

PPS tortured phrases on PDF text

Yes (screen, critique extract)

Yes (T4 dictionary + M1 graph)

GRIM / statcheck on manuscript inline stats

Partial (stdlib subset in critique)

Yes (full statcheck + GRIMMER/SPRITE)

Spreadsheet / table fabrication tests

No

Yes (A1/A3/A5, Benford, …)

Image duplication / splice forensics

No

Yes (F1–F6; raster supplements)

TIVA / P-curve / Carlisle

No

Yes

Pre-submission structure / refs / overlap pack

Yes (critique)

No

Zotero write-back / mirror restore

Yes

No

Complementary: Zotero plugins and built-ins

Paperful

  • Sites: GitHub · this repo

  • Fit: Research helper for a reference library: clean records, find missing PDFs, summarise papers, keep a platform-agnostic mirror. Routed OA stack (including CORE with API key), campus EZProxy, optional Scholar cookies, opt-in Sci-Hub, user playbooks for grey literature and field-specific hosts, and an opt-in AI browser after scripted lanes fail. Collection-mirrored out/ tree (snapshot / restore, including snapshot --pdfs all for PDFs already in Zotero, manifest.jsonl), attach on Zotero 10+. Identifier verify (Crossref/OpenAlex/PubMed) + lint + fix-metadata on disk, then adapter write-back. dedupe writes a review pack and, with --apply, merges the extra parent’s PDF, notes, and better fields onto the keeper, then trashes that parent (DOI, then title+year; Zotero only). gaps counts missing PDFs and DOIs. Optional local summarize / synthesize and recover (last run lane after Scholar / EZProxy / htmlpdf fail, or recover --item; text layer from the PDF or from paperful ocr; not a chat agent). Zotero is the well-tested adapter. Mendeley (REST) and EndNote (read-only database plus an import bundle) are seeking testers.

  • With others: paperful attachments for a CLI report (and flagged repairs from out/); StorScan or Attanger when you want the work inside Zotero, including incoming downloads; zotero-agent for disk GC or a two-up scan split; zotero-mcp when the work is a conversation.

  • Not a substitute for: In-app plugin UX, .bib hygiene tools, or two-up scan splitting.

Zotero built-in

  • Sites: Zotero · custom PDF resolvers

  • Fit: Per-item Find Available PDF, identifier lookup, duplicate merge UI, File → Export PDFs (a one-off dump of files already on disk, not a restore ledger).

  • With Paperful: Built-in capture on save; Paperful for batch backfill and for snapshot / restore when Zotero file storage is the wrong warehouse.

zotero-zotadata

  • Sites: GitHub

  • Fit: Plugin combining attachment validation, multi-source PDF retrieval, and metadata updates.

  • With Paperful: Overlapping PDF mission; zotodata stays in Zotero.

StorScan

  • Sites: GitHub

  • Fit: Attachment operations — scan, fix misplaced linked files, merge duplicate files.

  • With Paperful: Paperful attachments reports broken links, ghosts, and same-file duplicates, and can repair them from out/ when you pass a flag and --apply. StorScan is the in-app scan.

ZotMeta

  • Sites: GitHub

  • Fit: Bulk metadata from DOI, ISBN, arXiv; PDF cache identifier extraction.

  • With Paperful: Overlaps lint / fix-metadata; ZotMeta is in-app.

Attanger and ZotMoov

  • Sites: Attanger · ZotMoov

  • Fit: ZotFile-era workflows on Zotero 7+ — match downloads, rename/move linked files.

  • With Paperful: Incoming-file organisation and tablet send/get. paperful attachments --rename and --link cover the mirror-stem rename and a stored-to-linked cutover under out/, and they stay off until --apply.

zotero-agent

  • Sites: GitHub

  • Fit: Local-first CLI and MCP via bridge plugin — search, dedupe, merge, enrich, pdf-fetch, PDF notes, and pdf-prep OCR (OCRmyPDF) for scans. export writes bibliography formats, not a PDF folder tree.

  • With Paperful: Agent for hygiene and merge; Paperful for collection runs with a manifest, a disk text layer (ocr), EZProxy, grey-lit playbooks, and a folder you can restore from. zotero-agent pdf-prep still owns two-up page splits.

zotero-mcp ecosystem

  • Sites: richardjlyon/zotero-mcp · cookjohn/zotero-mcp · mcp-zotero

  • Fit: LLM-facing search, PDF text, create items, sometimes Unpaywall attach. Some forks OCR scans (Docling) or write reading notes from chat.

  • With Paperful: MCP when a person is in the loop; Paperful for unattended collection runs, the disk ledger, ocr on image PDFs, and batch summarize / synthesize.

Valency MCP Bond and Hub

  • Sites: valency.io

  • Fit: MCP Bond — hosted semantic access to a very large public paper/preprint corpus (search, citation graph, authors, trends, export) for LLM clients. Hub — open publishing, versioning, and review; out of scope for Paperful’s library/mirror jobs.

  • With Paperful: Bond when the question is “what does the literature say?” across hundreds of millions of works. Paperful when the question is “fill, fix, mirror, and reason over my collection” — run, snapshot, snowball / authorwatch to grow it, ask / precedent / refs_gap / gaps / dry-run snowball_search / snowball trends / export-proposals (and optional paperful mcp with the same JSON envelopes) over the on-disk index. Typical stack: Bond (or similar) for discovery → snowball / ingest-dois / manual save → Paperful for PDFs and disk-first hygiene.

  • Not a substitute for: EZProxy-backed fetch, grey-literature playbooks, collection-shaped out/ restore, or unattended batch attach.

Edison Scientific

  • Sites: docs.edisonscientific.com · platform.edisonscientific.com · FutureHouse spinout

  • Fit: Hosted agents that do research, not a Zotero helper. Split the jobs in copy: Literature (PaperQA3) is cited review over a licensed public corpus (docs 175M+; a 2026 blog says 150M+ with figures). Precedent (formerly Has Anyone) is novelty, not a second Literature call. Kosmos is UI-only (not on the public task API): ~1,500 papers and ~42k lines of code per run, shared world model, traces; the Kosmos paper’s expert eval is weaker on interpretation than on quotes. Academic credits moved between 2025–2026 posts — treat pricing as policy, not SLA. Analysis / Molecules are lab/chem. Local OSS cousin is PaperQA2 (paper-qa); hosted Literature is not “chat with my Zotero.”

  • With Paperful: Edison when the question is a public-corpus review, a novelty search, or a Kosmos/Analysis run on a dataset. Paperful when the question is “fill, fix, mirror, and Ask this library.” Local analogue of Precedent is paperful precedent (held PDFs + OpenAlex metadata, stamped held vs public-metadata-only; never “nobody has done this”). Typical stack: Edison Literature or Precedent → snowball / ingest-dois → run / Wanted → local ask. Do not treat their workspace as the quiet mirror. Do not borrow their paper-scraper lane as a default (their own docs flag publisher ToS grey).

  • Not a substitute for: missing-PDF fetch, EZProxy / grey playbooks, snapshot / restore, collection-scoped RAG that never leaves the machine, or unattended attach.

  • Nearby (not Edison): Google Co-Scientist / Labs Hypothesis Generation (hypotheses, not manuscripts); Sakana AI Scientist (workshop-grade auto-papers; authors documented citation errors); OpenAI / Gemini / Perplexity Deep Research (web memos; vendors warn about hallucinated cites).

  • Roadmap borrow (not a clone): ROADMAP — Hosted AI scientists and v2 factory packs.

Systematic-review screening

  • Sites: Covidence · Rayyan · DistillerSR · ASReview (local OSS)

  • Fit: Import → dual screen → extraction forms → audit / PRISMA counts. Usually no discovery corpus and no campus PDF fetch.

  • With Paperful: they screen; Paperful fills PDFs and writes Zotero. Discover Keep/Skip is hop triage. Interop (export queue / import included.csv) is the SnowBallSLR lesson, not vendoring Covidence.

Manuscript assist

  • Sites: Jenni · Paperpal · SciSpace AI Writer · Writefull in Overleaf

  • Fit: Draft + cite from library ∪ a 200–300M index. Mitigations are mostly existence checks (real DOIs), not “this sentence is supported by that passage.” Elsevier’s 2026 author factsheet still wants a human intellectual contribution and forbids fabricating references.

  • With Paperful: do not become Jenni. Extra cites belong in Zotero before \cite (stuff). Living reviews without a model: Better BibTeX → Quarto.

PaperQA2 (local OSS)

  • Sites: future-house/paper-qa

  • Fit: Cited Q&A over PDFs you supply. This is the cloneable pattern (chunk → retrieve → refuse if missing). Hosted PaperQA3 / Edison Literature is a different product (licensed corpus + figures).

  • With Paperful: Paperful already owns get-PDF + mirror + Zotero write-back. Ask --ask-mode advanced is the score-then-answer path on the local index (not a vendored paper-qa, no 175M corpus). Do not wrap their Zotero extra (maintainers note it has been a while since they tested it).

ZotFile (legacy)

  • Sites: zotfile.com

  • Status: Not compatible with Zotero 7+. Use Attanger / ZotMoov.

BibTeX, Mendeley, and scripts

bibcite and bibtex-tidy

  • Sites: bibcite · bibtex-tidy

  • Fit: Resolve papers to canonical BibTeX, dedupe, tidy, upgrade preprints.

  • With Paperful: Orthogonal until a bib-file LibraryBackend exists.

bibmanager

  • Sites: Read the Docs

  • Fit: Unified BibTeX database, ADS integration, optional PDF fetch from ADS.

Mendeley

  • Fit: Reference Manager Duplicates smart collection; Desktop merge UI. Paperful can set manager = "mendeley" (REST, OAuth). That adapter is seeking testers. Zotero is the well-tested path.

paperscraper

  • Sites: GitHub

  • Fit: Python metadata + save_pdf from DOIs; publisher TDM API keys.

  • With Paperful: paperscraper when the input is a DOI list, not a collection tree.

Choosing in one glance

Need PDFs already in Zotero copied into a collection-shaped tree, with a record per item?
  → paperful snapshot --pdfs all

Need a one-off dump of those PDFs from the Zotero UI?
  → File → Export PDFs

Need tablet send/get after ZotFile died?
  → ZotMoov custom menus (+ Attanger)

Need a text layer on a scanned PDF?
  → paperful ocr. Two-up books: zotero-agent `pdf-prep`

Need LLM to fix one paper’s metadata while writing?
  → zotero-mcp

Need a grounded summary of many text-layer PDFs, then one collection review?
  → paperful summarize, then synthesize

Need reproducible .bib for a paper submission?
  → bibcite fix / bibtex-tidy

Chat agent needs the whole public literature (graph, trends, author intel), not just Zotero?
  → Valency MCP Bond (complement Paperful for *your* PDFs and mirror)

Need a hosted cited review, “has anyone done this?”, or Kosmos on a dataset?
  → Edison Scientific. Keep Paperful for PDFs, the disk ledger, and Ask over held papers

Need Elicit/Consensus tables over the public record?
  → those UIs, then Paperful for PDFs you will own

Need dual-review SR screening?
  → Covidence / Rayyan / ASReview

Need a web Deep Research memo?
  → OpenAI / Gemini / Perplexity. Not Zotero

Need a manuscript draft?
  → Jenni / Paperpal / Overleaf. Paperful does not write the paper

Need local cited Q&A on a folder, no Zotero?
  → PaperQA2. Paperful when the folder should also restore into a manager

Already have a reading list on disk and want gaps vs what your PDFs cite?
  → paperful refs gap (CLI or MCP refs_gap)