SYSTEM ONLINE

FOR CLAUDE CODE

MIT LICENSE

🤖

researchers have poked Scholark-1 3,255 times

SCHOLARK-1

Your AI assistant is brilliant at reasoning. But it invents paper titles, fabricates DOIs, and hallucinates citation counts. Scholark-1 connects it to real academic databases so every reference it gives you actually exists.

SYSTEM MANIFEST

WHAT

An MCP server that gives your AI agent direct access to real academic papers, real metadata, real BibTeX, and real open access PDFs. It now reads your own PDF stash too. 10 tools, 5 data sources, zero hallucinated references.

HOW

Searches 4 databases in parallel. Deduplicates results by DOI. Falls back automatically when a source is down. Caches lookups in SQLite. Every result cites its source.

SETUP

One command to install. Zero API keys required. Built for Claude Code, where your AI already has access to your project, your draft, and your refs.bib.

COST

Free. Forever. MIT licensed. All underlying APIs are free for research use. No paid tier, no usage limits, no tracking.

TRUST

Every result states its source. 4 databases cross-checked and deduplicated by DOI. Retracted papers carry a loud alert, preprints are labeled as not peer reviewed, and uncertainty is flagged, never hidden. Your AI stops making things up.

YOUR PDFS

Point it at your messy pdf-dump folder. Every PDF gets matched to its real DOI, so full-text search hits come back citable with filename, page number, and BibTeX. Retracted papers hiding in your own library get flagged. Other PDF-RAG tools cite "file.pdf, p. 12". Scholark-1 cites the actual paper.

REPRODUCIBLE

Every search and every paper considered is logged to .scholark-1/session-log.md in your project. Your literature trail, gitignored, in your repo. Answer reviewers, reflect on your own bias, and never wonder "did I look at that paper?" again.

// ACTIVE TOOLS

search_papers

Search 4 databases in parallel. Results deduplicated by DOI. Source attribution on every paper.

fetch_paper_details

Deep metadata with automatic fallback across 4 sources. Every DOI is cross-checked against OpenAlex's retraction flag, fresh on every call. Cached for speed.

search_by_topic

Topic search with year range filtering across 3 databases in parallel, deduplicated by DOI. Find what was published on X between 2020 and 2025.

doi_to_bibtex

Any DOI to a BibTeX entry. Paste a DOI, get a .bib-ready citation. Cached for 90 days.

find_open_access

Find free, legal PDFs via Unpaywall. See OA status, version, license, and download links.

get_citation_context

The actual sentences where other papers cite a work. See how a finding was received, criticized, or extended.

index_local_pdfs

Index a local folder of PDFs. Optional, since search auto-indexes, but handy for a warm-up run or detailed per-file stats. Each file is matched to its DOI via printed DOI, arXiv ID, or Crossref title lookup. Incremental, so re-runs only touch new files. Retracted papers get flagged on the spot.

search_local_pdfs

Full-text search over your own library. Keeps the index in sync automatically, new PDFs in the folder just show up. Hits come back with filename, page number, and DOI, ready for doi_to_bibtex. Online search results also show when you already have the paper on disk.

resolve_local_pdf

Attach a DOI by hand to the stubborn files. After that they carry full citation info and retraction checks like everything else.

enable_semantic_search

One chat command installs a local embedding model (~100MB, one time) and upgrades local search to hybrid keyword + meaning matching. Optional, off by default. Set up once, new PDFs get embedded automatically from then on.

// DATA SOURCES

SEMANTIC SCHOLAR

OPENALEX

CROSSREF

EUROPE PMC

UNPAYWALL

YOUR PDF FOLDER

// .scholark-1/session-log.md

# Auto-maintained in your project root. A trail of what you actually considered.
- 2026-05-02 14:31:07 | search_papers | "LLM hallucination detection 2023-2025": 8 papers (10.1145/3491102.3517582, 10.1109/...)
- 2026-05-02 14:35:42 | fetch_paper_details | 10.1145/3491102.3517582: full metadata retrieved
- 2026-05-02 14:42:18 | doi_to_bibtex | 10.1145/3491102.3517582: BibTeX returned, added to refs.bib
- 2026-05-02 14:50:33 | search_by_topic | "estimation statistics in HCI" (2023-2025): 5 papers
- 2026-05-02 15:02:51 | get_citation_context | 10.1145/3491102.3517582: 12 citing papers, 4 critical responses

# Reviewer 2: "you missed paper X". You: opens session-log.md, finds the entry, replies with confidence.

// 2-STEP SETUP

# 1. Add to any project
claude mcp add -s project scholark-1 -- uvx --from git+https://github.com/SHosio/scholark-1 scholark-1

# 2. Ask your AI to search for papers. That's it.
"Find recent papers on retrieval-augmented generation for scientific literature"

# 3. Optional: point it at your own PDFs
echo 'SCHOLARK_PDF_DIR=~/Documents/pdf-dump' >> .env

// HOW TO USE

Each prompt below maps to one or two tool calls. No orchestration needed. Just ask your AI:

"Get the BibTeX for 10.1145/3491102.3517582 and add it to refs.bib"

"Find an open-access PDF for 10.1145/3411764.3445428"

"Read related-work.tex and find 5 recent papers from 2023–2025 I should cite"

"Show me how 10.1145/3411764.3445428 has been cited. Any critical responses?"

"Look up every DOI in refs.bib and flag any that don't resolve"

"Find survey papers on LLM hallucination detection from the last two years"

"Index my pdf-dump and tell me if any of my PDFs are retracted"

"Search my local PDFs for construct validity definitions, quote with page numbers"

Want orchestrated workflows?

Scholark-1 gives your AI six stateless primitives. For multi-step research workflows (read my draft, find blind spots, generate a literature gap report, brainstorm with a divergent ideator and an adversarial critic), pair it with Scholark, the companion Claude Code plugin. Two researcher-shaped agents and six skills built on top of scholark-1's tools.

Free and open source. MIT licensed.
Give your AI agent the academic literature it's been missing.

Install scholark-1

No API keys required to start. Core tools work immediately.
Optional config unlocks open access PDFs, higher rate limits, and your local PDF library. See README.