Run refcheck

From the root of your LaTeX project:

refcheck run

The command extracts the current citations, starts a temporary API bound to your computer, and opens a private session on this website. Your LaTeX source, annotations, PDFs, and Zotero database remain local.

If the connection prompt appears, allow refcheck.ch to connect to software running on your computer. Closing the terminal stops the local session.

Configure your project

refcheck init creates refcheck.toml. A minimal working template is:

main_tex = "main.tex"
bib_file = "references.bib"

zotero_db = "~/Zotero/zotero.sqlite"
zotero_storage = "~/Zotero/storage"
work_dir = ".refcheck"

# Optional local AI suggestions
llm_base_url = "http://127.0.0.1:1234/v1"
llm_model = "google/gemma-4-e4b"

# External literature search
scite_mcp_url = "https://api.scite.ai/mcp"

Paths are resolved from the project root except for ~ paths. If your config predates scite support, add scite_mcp_url manually; refcheck init does not overwrite an existing file.

Use local AI suggestions

Start LM Studio's local server at http://127.0.0.1:1234, load the model configured in refcheck.toml, and click Suggest with AI on a citation. refcheck scans the local PDF, verifies candidate passages against the PDF text, and lets you open the matching page in Zotero. Your PDF text stays on your computer.

Search for a better reference

Click Connect scite on a claim and complete the OAuth sign-in to enable Search using external services. refcheck uses PKCE and keeps the resulting tokens in the local project work directory. The search sends only the selected claim sentence—not your project files or PDFs.

If refcheck reports a TLS certificate error, the endpoint is configured but the Python installation or an HTTPS-inspecting proxy on that machine does not trust the certificate chain. This is different from the targeted scite_mcp_url configuration error shown for an empty or malformed URL.