Catch fabricated DOIs, retracted papers, and misquoted text before publishing.
发稿交审前,自动揪出假 DOI、已撤稿论文与错引原文。
claimcite verifies every DOI, PMID, arXiv ID, and web link in Markdown or plain text directly against doi.org, Crossref, OpenAlex, PubMed, and arXiv — no LLM API keys, no document uploads.
直接基于 doi.org、Crossref、OpenAlex、PubMed 与 arXiv 公开注册表逐条核查 Markdown 或纯文本中的引用。零第三方运行时依赖,不需要大模型 Key,也不上传整篇文稿。
uv tool install claimcite
doi:10.1007/s10654-019-00524-3
doi:10.1016/s0140-6736(10)61388-8
doi:10.1001/jama.2022.99999
pmid:33237064
What it checks
核心核查能力
Deterministic checks backed by authoritative registries — no LLM guessing.
全部基于公开权威文献接口做确定性核对,不用大模型猜答案。
Identifier & OpenAlex Batching
多源标识符与 OpenAlex 批量查询
Verifies DOIs via doi.org, PubMed, Crossref, and 50-DOI OpenAlex batches (including DataCite DOIs), plus PMIDs and arXiv Atom queries.
支持 doi.org、PubMed、Crossref 及最多 50 条一批的 OpenAlex 检索(覆盖 DataCite 与跨学科 DOI),以及 PMID 与 arXiv 批量接口。
Retraction Detection
撤稿检测(不误伤撤稿声明)
Flags retracted papers via PubMed Retracted Publication / RetractionIn, Crossref updated-by / relation.is-retracted-by, and OpenAlex without flagging retraction notices themselves.
结合 PubMed、Crossref 与 OpenAlex 的撤稿关系字段标出被撤论文,同时避免把撤稿声明本身误判为被撤稿件。
Title, Year & Quote Match
题名、年份与引号原文比对
Checks title similarity (stripping RETRACTED: prefixes), publication year (±1 tolerance), and verbatim quotes against abstracts (including reconstructed OpenAlex inverted indexes), open-access full text (PMC / Europe PMC), or web pages.
比对题名一致性、年份偏差(±1 年容差),以及引号原话是否出现在摘要(含 OpenAlex 倒排索引还原摘要)或网页中;未命中时按需核查开放获取全文(PMC / Europe PMC)。
Numbered Audit & Title Search
编号一致性与无 DOI 文献检索
Flags dangling body markers ([5], [^1]) and uncited bibliography entries with line numbers, and searches unlinked APA/Harvard/MLA/Vancouver/GB/T 7714 titles in Crossref and PubMed.
报出正文引用却缺失条目的 [5] / [^1] 与从未被引用的尾注/参考文献,并自动识别中英文格式,按标题检索核实无 DOI 文献。
Study Type Signals
研究类型标注
Surfaces PubMed publication types and MeSH tags (Systematic review / meta-analysis, RCT, observational, animal study) directly in reports.
自动提取 PubMed 文献类型与 MeSH 标签(系统综述/Meta、随机对照试验、观察性研究、动物实验等)。
30-Day Local Cache
30 天本地缓存
Successful lookups are cached locally for 30 days under XDG_CACHE_HOME. Lookup failures are never cached. Pass --no-cache to bypass.
成功查询结果本地缓存 30 天(失败请求绝不缓存),重复核对毫秒级完成;也可用 --no-cache 强制刷新。
Install, MCP Server & CI
快速上手:CLI、MCP 与 CI
Run from the terminal, wire into Claude / Cursor / Codex via stdio MCP, or gate pull requests.
既可以命令行单次检查,也能作为零依赖 stdio MCP 接入 AI 编辑器,或放进 GitHub Actions 卡口。
pipx install claimcite
claimcite check article.md
claimcite check article.md --lang zh
claimcite check article.md --format json
{
"mcpServers": {
"claimcite": {
"command": "uvx",
"args": ["claimcite", "mcp"]
}
}
}
- uses: wasabihu/claimcite@v0.6.0
with:
files: "docs/**/*.md"
fail-on: error
repos:
- repo: https://github.com/wasabihu/claimcite
rev: v0.6.0
hooks:
- id: claimcite
Honest Scope & Boundaries
能力边界说明
- No unquoted paraphrase guessing: Only explicit identifiers, URLs, reference-list entries, and verbatim/near-verbatim quoted spans are checked.
- 不猜没有引号的自由改写:只核查明确写出标识符/链接的引用,以及引号内的原话比对,不擅自对未加引号的改写下结论。
- Not a semantic peer reviewer or medical advice: claimcite confirms that sources exist and match what the text says about them; it does not evaluate scientific validity.
- 不是语义审稿人,也不构成医疗建议:只确认“文献存不存在、信息对不对得上、是否撤稿”,不评判研究方法本身的优劣。
- Sister project: Building an audited Python agent runtime? See cairnkit, which includes a citation-checking approval workflow example using claimcite.
- 联动项目:如果你正在写需要人工审批边界的 Python Agent 后端,可参考姐妹项目 cairnkit(内含对接 claimcite 的引用核查审批示例)。