Bibverify MCP
BibTeX verification against Crossref, OpenAlex, PubMed and arXiv: a CLI, Python API and MCP server to flag wrong references
- 98 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The tool calls scholarly database APIs and sends them identifiers and titles of works
- The `--apply` mode rewrites the `.bib` file, although it makes a backup
Install
- Works in
Manual install
uvx bibverify --versionRuns without a permanent install; for a persistent one use `uv tool install bibverify`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The tool calls scholarly database APIs and sends them identifiers and titles of works
- The `--apply` mode rewrites the `.bib` file, although it makes a backup
This is third-party code. Review the repository files before installing.
What it does
Bibverify checks entries in a `.bib` file against scholarly databases and shows whether the work exists and whether its metadata agrees. Lookup starts with exact DOI, PMID, PMCID or arXiv identifiers, then candidates are scored on title, authors, year, venue and pages. Sources include Crossref, OpenAlex, Semantic Scholar, PubMed, Europe PMC, CORE, DBLP, arXiv and bioRxiv. If a DOI resolves but the title differs materially, the entry gets an `identifier_conflict` status and stays visible among title-search results. Fixes are a separate step: `bibverify check references.bib --apply` first makes a byte-for-byte backup and leaves fields such as `abstract`, `keywords`, `file` and `note` alone when sources do not return them. Conflicting identifiers are never overwritten automatically. The MCP server starts with `bibverify mcp` over stdio or Streamable HTTP and offers four tools: `doi_to_bibtex`, `rank_lookup_sources`, `explain_update_diff` and `verify_bib_file`. By default it reads and writes only inside the configuration directory.
Who it is for. Paper and thesis authors, editors and researchers who ask Claude Code or Codex to check a bibliography for invented or inaccurate references.
Good fit when
- An agent generated a reference list and you need to know which works really exist
- You want to turn a DOI into a ready BibTeX entry with `bibverify doi`
- Before submitting a paper you need to catch mismatched years, authors or venues in a `.bib`
- The check runs in CI and needs a JSON report and stable exit codes
- You want to review changes as a diff before writing to the file
Not a fit when
- You need to confirm a paper's findings are true: the tool checks only that a record exists and metadata agree
- A work is missing from databases because it is new or internal: a missing record does not prove forgery
- You want a reference manager with PDF storage, for which Zotero fits better
Example request
Check references.bib against Crossref and PubMed, show entries with mismatches and do not change the file.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The tool judges whether a record can be found in the queried databases and whether metadata agree. Whether findings are true, data authentic or a venue reputable lies outside the check. Network access to the database APIs is required and some sources rate-limit requests; rate-limit, auth and network errors are separated from a genuine no-match in the report. Version notes in the README run ahead of the published release: some launch methods are marked available from version 0.4.0, so check the version with `uvx bibverify --version`. The `--apply` mode changes the file, so review the report with `--dry-run` first.
How to disable. Remove the server from the MCP configuration. An installed package is removed with `uv tool uninstall bibverify` or `pipx uninstall bibverify`.
FAQ
How do I connect bibverify to Claude Code?
Add a server with the command `bibverify` and the arguments `mcp --config config.json`. The `uvx bibverify` form works without installing, and Node clients can start it with `npx --yes @hylouis233/bibverify mcp`.
Does it change my .bib without asking?
No. A regular check writes reports only; edits require the `--apply` flag and a byte-for-byte backup. `--dry-run` writes no files.
Does bibverify prove a reference is fake?
No. A record missing from databases may mean the work is new, internal or formatted differently. The tool shows mismatches with evidence and leaves the decision to a person.
Which databases does bibverify use?
Crossref, OpenAlex, Semantic Scholar, PubMed, Europe PMC, CORE, DBLP, arXiv, bioRxiv and others. Successful responses are cached in SQLite with an expiry; failures are never cached.
Which tools does the MCP server provide?
Four: `doi_to_bibtex`, `rank_lookup_sources`, `explain_update_diff` and `verify_bib_file`. The server supports stdio and Streamable HTTP.
Discussion
No questions or reviews yet.
Related
MCP serversA local MCP server for arXiv literature work: section-level paper reads, BibTeX from metadata, topic watches
BibTeX Verifier checks .bib files against OpenAlex, CrossRef and DataCite and flags references with errors or invented data.
MCP server with 40 tools that checks quotes, numbers, DOIs and PRISMA counts in AI-written literature reviews. No model inside.
Academia MCP: scientific paper search
Academia MCP
An MCP server for scientific work: search arXiv and ACL Anthology, Semantic Scholar citations, PDF reading and LaTeX compilation
A Python CLI and library that turn DOIs, PMIDs, arXiv IDs, ISBNs and BibTeX drafts into auditable BibTeX or CSL-JSON
Scholar Search MCP connects a discovered paper to its references and later citations when preparing a literature review.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Bibverify (this card) | no | not required | 98 | |
| BibTeX Verifier | no | 144 | a BibTeX verification skill without a CLI or its own MCP server | |
| Zotero MCP | no | not required | 1,278 | agent access to a Zotero library; it does not check a .bib file |
README in short
The README shows that checking moves from identifiers to scoring on title, authors, year, venue and pages. It describes installation via uvx, uv tool, pipx, npm and a container, the `config.json` setup, JSON output, exit codes and an SQLite cache. A separate section covers MCP and generating ready settings for Codex, Claude and Cursor.