-
Notifications
You must be signed in to change notification settings - Fork 0
FAQ
The server is installed but does not appear in Claude Desktop. Quit Claude Desktop completely, then reopen it. Closing the window does not reload the server list. On macOS, make sure the icon is gone from the Dock (⌘Q).
Do I need to install Node.js first? Not for Claude Desktop — it ships its own. Node.js is only needed if you build from source or use these in Claude Code.
Do I need Python, pip, or uv?
No. All three servers are pure Node.js. If you find instructions telling you to run
install.sh or install a PyPI package, they belong to an older pack and no longer apply.
Where do I put the .env file?
There isn't one. No server here ever reads .env. In Claude Desktop the fields are in the
extension window; in Claude Code, use -e NAME=value with claude mcp add. scr-toolkit-nulis
has no fields at all.
The Scopus tools do not appear at all.
That is correct behaviour when SCOPUS_API_KEY is empty. The five Elsevier tools are not
registered without a key, rather than appearing and failing when called — tools/list
returns 16 instead of 21. Set the key, reinstall the extension, then restart Claude Desktop.
The key is set but Scopus refuses with 401 or 403.
Run nulis_elsevier_status — it reports whether the key is read and whether it is actually
accepted. The most common cause is accessing from outside the campus network. Ask your
librarian or licence admin for an insttoken and set ELSEVIER_INSTTOKEN.
How much quota do I have left?
nulis_elsevier_status reports it. Run it before starting a systematic search, not after the
quota runs out mid-way.
nulis_get_open_access_pdf finds nothing even though the article clearly exists.
The article genuinely has no legal open-access copy. This tool only points at legitimately
open copies; it does not look for pirated ones. Note that a missing CONTACT_EMAIL is not
the cause — without it the lookup still runs through OpenAlex. Setting the email adds the
Unpaywall route, which also reports the copy's licence.
Can I trust the metadata from search_*?
For citation purposes, verify with nulis_get_paper_by_doi. Search results are useful for
finding, but what binds is what Crossref has registered. A citation that merely looks
plausible is the signature pattern of invented references.
Where are the PDFs saved?
DOWNLOAD_DIR if set, then ~/Downloads, then the system temp folder. Run nulis_server_status
to see which folder is actually in use right now.
"Zotero unreachable" even though the app is open. Open Zotero → Settings → Advanced → tick "Allow other applications on this computer to communicate with Zotero". This is the step most often missed.
Full text is empty even though the item clearly has a PDF. The server reads Zotero's own full-text index rather than re-parsing the PDF. If Zotero has not finished indexing that file, the result is empty. Right-click the item → Reindex Item, then try again.
Can this server modify my library?
No. Every call it makes is a GET; there is no write path in the code at all.
The BibTeX citation keys differ from the ones I use.
nulis_zotero_export_bibtex uses Zotero's own data. If you use Better BibTeX, the keys you
manage there are handled by that plugin and will not always match. Check before pasting into
a LaTeX manuscript already underway.
What are M6 Langkah 2, M5 L3, M9 L8 in the tool descriptions?
Module steps in the Alur SLR AI course — this tool is a companion to those modules and
deliberately stays one. Inside the course, those references tell participants exactly when a
tool is used. Outside it, the table on the scr-toolkit page is enough; the
tools require no course to run.
Does this decide which studies are included? No. None of its nine tools decides eligibility. It checks, counts, matches, and retrieves — the decision stays with the researcher.
The result is TIDAK_DAPAT_DIPERIKSA rather than match or mismatch.
That is the correct answer, not a failure. It means "cannot be checked". Without
pdftotext, the built-in extractor may only prove a match and never deny one — measured
over 432 quotations, of those it missed, 18 of 23 were genuinely present. Install poppler if
you need accusing verdicts. The same answer is returned for scanned PDFs, recognised by text
density below 800 characters per page.
Do I have to install poppler?
No. Everything still runs without it; only the authority of the check is reduced, as above.
On macOS: brew install poppler.
My PDFs were downloaded by another MCP (scholar/Zotero) — safe to use directly?
No. Files from elsewhere are not named SCR[ID]_, so filename-based checks cannot see them
— even though those are precisely the ones needing verification. One download once returned
a valid PDF containing an entirely different article. Run nulis_pdf_integrity →
nulis_pdf_match_records → nulis_pdf_verify_record first.
nulis_xlsx_write deleted my other sheets.
It rewrites the whole file and is not a cell editor. Include every sheet you want to keep,
or write to a new file.
Is any of my data sent anywhere?
No telemetry and no relay server. scholar-nulis calls public APIs directly from your computer;
zotero-nulis in local mode only talks to the Zotero app on localhost; scr-toolkit-nulis only
touches the files you point it at, plus Unpaywall and publisher pages when you ask it to
download.
Is anything still only in Indonesian?
Two things. The tool descriptions inside manifest.json — that text is read by the model,
not by you, but it is also what Claude Desktop prints in the extension window, so that
screen reads as Indonesian. And scr-toolkit-nulis's own README, which records its measured
limits; the scr-toolkit page covers the same ground in English. Everything
else on this site has an English page, and the Indonesian originals live in
docs/id/.
What is the relationship with the skills repo?
This repo is the tools; nulis-not-just-writing/skills
is the procedures. They complement each other and neither requires the other — those skills
work fully without these servers, and these servers are useful without the skills.
What is the licence?
CC BY-NC 4.0, the same as the skills repo. Free to use, copy, adapt, and share for
non-commercial purposes with attribution. Researchers, students, lecturers, and educational
institutions need no permission at all; just credit the source. Commercial use requires
separate permission from the rights holder.
If the code is CC BY-NC, what about the bundled MIT libraries?
The MIT licence continues to govern those libraries' own code; CC BY-NC applies to the
original work only. MIT permits a combined work to be distributed under different terms as
long as its copyright notice is retained — which is why NOTICE.md sits inside every
.mcpb, not just in the repo. A file downloaded on its own does not carry the repo with it,
so the attribution has to be inside the bundle. build-mcpb.sh refuses to produce a bundle
without one.
scr-toolkit-nulis bundles no third-party code at all — zero npm dependencies — so there is no
other copyright to reproduce for it.
← Back · Installation · scholar · zotero · scr-toolkit
Knowledge unshared dies. Knowledge shared keeps living.
It grows in hands you will never meet and is carried on in work you will never read — and what never stops living never stops returning to you.
Mubaroq ADB · Akademi Digital Bandung | RPI Institute · mubaroq@digitalbdg.ac.id
Mirror of docs/ — do not edit here; edit in the repository. ·
Cerminan docs/, jangan disunting di sini. · CC BY-NC 4.0
Start
Servers
More