Hylouis233/bibverify

View on GitHub ↗

Cross-platform BibTeX verification for researchers and AI assistants — CLI, Python API, MCP server, and explainable multi-source matching.

31 ★0 forksPythonUpdated 20d ago

What you need to know

BibTeX metadata verification tool and MCP server: it starts from exact DOI, PMID, PMCID, or arXiv identifiers, then scores candidates from title, author, year, venue, and pagination evidence across Crossref, OpenAlex, Semantic Scholar, PubMed, Europe PMC, CORE, DBLP, arXiv, bioRxiv, and more — verifying bibliographic existence and metadata consistency without destructive edits.

Install

uvx bibverify
uv tool install bibverify
pipx install bibverify
docker run --rm ghcr.io/hylouis233/bibverify:0.4.0 --version

Usage

  • Start with identifier-first lookups: bibverify doi 10.1038/nature12373 or bibverify check references.bib --config config.json.
  • Inspect results without writing anything via --dry-run --json; only --apply modifies the source, and only after creating a byte-for-byte backup.
  • Start the MCP server with bibverify mcp [--transport stdio|streamable-http]; it treats the config directory as workspace root and blocks reads/writes outside it.
  • Generate per-client setup files with bibverify agent init --target codex|claude|cursor.

Key features

  • Multi-signal matching with explainable results (a resolvable DOI with a materially different title becomes identifier_conflict instead of being hidden).
  • Structured provider outcomes distinguishing genuine no-match from rate limiting, auth, network, and parse failures.
  • Non-destructive merging preserves abstract, keywords, file, note, and custom fields; conflicting persistent identifiers are never overwritten automatically.
  • Retry/backoff connection pool with Retry-After handling; expiring SQLite cache (failures are never cached).
  • MCP server on the official SDK supporting stdio and Streamable HTTP with tools doi_to_bibtex, rank_lookup_sources, explain_update_diff, verify_bib_file.
  • Cross-platform (Windows/macOS/Linux) with JSON output and stable exit codes.

Best for

Researchers, editors, automation, and AI assistants that need to check whether references exist and whether their metadata agrees with authoritative sources.

Caveats

  • Current published release is v0.3.0; the npm launcher, native builds, and GHCR image are planned for v0.4.0 and not yet published (Homebrew/Scoop/WinGet catalogs not live).
  • Windows ARM64 native builds are not published yet (npm uses the x64 build under Windows 11 emulation).
  • --apply edits the source .bib after a backup; the default command only writes proposals.
  • Missing database records do not prove fabrication, and verification does not judge venue reputation.
  • bioRxiv title search is not supported (exact 10.1101/... DOI only).
Platforms: macOS · Linux · WindowsClients: Codex · Claude · Cursor

Reviewed 2026-08-18

Topics

Stars
31★
Forks
0
Language
Python
License
MIT
Created
2025-10-11
Last push
2026-08-15