UniArticles MCP Server
Offers capabilities to search papers, list recent papers, retrieve paper metadata, and download PDFs from ArXiv.
Enables access to Elsevier's Scopus and ScienceDirect APIs for academic literature search and retrieval, requiring an Elsevier API key.
Provides experimental search of paper metadata by title from Google Scholar via Paperscraper.
Allows searching for papers from PubMed via the Paperscraper API.
Provides tools for searching documents, retrieving abstract details, author profiles, author search, and API quota checking from the Scopus database.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@UniArticles MCP Server search arXiv for recent papers on transformer models"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
UniArticles MCP Server
总览
亿文通(UniArticles)是一个实现了模型上下文协议 (MCP) 的统一学术文献检索服务器。截至 v3.5.0,它把 9 个数据源、29 个工具——Scopus、ScienceDirect、arXiv、PubMed、Crossref、Europe PMC、DOAJ、OpenAIRE、CORE——统一到同一套标准化接口下,供 LLM 客户端(Codex Desktop、Cherry Studio、Claude Desktop 等)调用。
Related MCP server: trove-mcp
功能特性
统一接口: 所有数据源使用统一的返回结构。
多源支持: 支持9个不同领域的、包括OA与非OA的数据源查询。
标准化返回: 一致的 JSON 结构 (
ok,source,query,count,items,error)。
当前支持的文献数据源
亿文通将以下 9 个数据源统一到同一套 MCP 接口下,全部返回相同的归一化 JSON 结构,且全部默认启用,共提供 29 个工具。除 arXiv 通过官方 arxiv Python 包封装外,其余每个数据源都是通过 httpx 直连该服务商的官方 REST API。
数据源 | 覆盖范围 | 接入方式 | API Key |
Scopus | Elsevier 精选的摘要与引文数据库,覆盖自然科学、社会科学、艺术与人文。 | Elsevier REST API( | 必需 —— |
ScienceDirect | Elsevier 的同行评审期刊与图书全文平台。 | Elsevier REST API( | 必需 —— |
arXiv | 物理、数学、计算机科学、定量生物、经济学等领域的开放预印本。 | 官方 | 无需 |
PubMed | 美国国立医学图书馆(NCBI)收录的生物医学与生命科学文献。 | NCBI Entrez E-utilities REST API( | 可选 —— |
Crossref | 覆盖所有学科的 DOI 注册元数据。 | Crossref REST API( | 无需 |
Europe PMC | EBI 的生命科学文献聚合库(区别于 NCBI PubMed),含 PMC 全文。 | Europe PMC REST API( | 无需 |
DOAJ | 开放获取期刊目录(Directory of Open Access Journals)中的同行评审文章。 | DOAJ REST API( | 无需 |
OpenAIRE | 欧洲开放科学研究成果聚合库。 | OpenAIRE REST API( | 无需 |
CORE | 汇聚全球仓储与期刊的开放获取论文聚合库;9 个数据源中唯一提供分布统计(年 / 出版社 / 学科等 facet)与机构库画像的源。 | CORE v3 REST API( | 可选 —— |
⚠️ API 密钥说明
本服务器提供的每一个工具、每一个数据源,都可以用您以个人身份申请的 API Key 访问,或者根本不需要 Key——没有任何一项需要机构订阅。 9 个数据源中有 5 个完全不需要 Key,需要 Key 的 4 个也都支持个人免费申请。
密钥 | 用于 | 申请方式 |
| Scopus + ScienceDirect(8 个工具) | 在 Elsevier Developer Portal 注册免费个人账号后创建 API Key。基础级、非商业性质的 Key 即可满足本服务器全部 Elsevier 相关工具,不需要机构订阅,也不需要 Insttoken(已用真实的非商业 Key 逐一实测验证)。Scopus 是 Elsevier 旗下数据库,因此在密钥作用域允许的前提下,同一把 Key 也可用于其他 Elsevier API 服务。 |
| CORE(9 个工具) | 前往 core.ac.uk/services/api#form 申请。可选——不配置也能用,但额度低得多:未认证档为 100 tokens/天、10 次/分钟,且官方不提供 |
| PubMed(4 个工具) | 先在 ncbi.nlm.nih.gov 登录 NCBI 账号,再到 NCBI 账号设置页 申请。可选——不配置也能用;配置后仅将限速从 3 请求/秒提升到 10 请求/秒。 |
完全不需要 API Key 的数据源:arXiv、Crossref、Europe PMC、DOAJ、OpenAIRE。
注意:即使一个 Key 都不配置,服务器仍会注册并暴露全部 29 个工具——只有对需要 Key 的数据源的调用会失败,而且是以清晰的错误信息失败,不会从工具列表中悄悄消失。
安装与使用
方法一:直接集成到 LLM 客户端(推荐)
适用于 Cherry Studio、LM Studio、Claude Desktop、Trae 等。
本项目已发布至 PyPI,您无需下载完整项目源码,直接通过配置即可使用。
由于上述 LLM 客户端通常内置了 Python 和 uv 环境,您无需额外下载,只需在客户端的 MCP 配置文件(如 claude_desktop_config.json)中添加以下内容即可:
{
"mcpServers": {
"uniarticles-mcp-server": {
"command": "uvx",
"args": [
"--refresh",
"uniarticles-mcp"
],
"env": {
"ELSEVIER_API_KEY": "your_elsevier_api_key_here",
"NCBI_API_KEY": "your_ncbi_api_key_here",
"CORE_API_KEY": "your_core_api_key_here"
}
}
}
}关于
env字段:只有ELSEVIER_API_KEY是必需的(用于 Scopus / ScienceDirect),其余全部为可选项——如果您没有某个 Key,请整行删除(JSON 不支持注释,且删除后剩下的最后一行末尾不能带逗号)。各可选字段说明:
如果您不希望每次重启时强制刷新缓存包,则改为添加以下内容:(但这会导致包更新时您需要对包进行手动更新)
{
"mcpServers": {
"uniarticles-mcp-server": {
"command": "uvx",
"args": [
"uniarticles-mcp"
],
"env": {
"ELSEVIER_API_KEY": "your_elsevier_api_key_here",
"NCBI_API_KEY": "your_ncbi_api_key_here",
"CORE_API_KEY": "your_core_api_key_here"
}
}
}
}📖 如果您在该方法下遇见了任何问题,详见:傻瓜式配置攻略
如果您在启动服务时遇到 “MCP error -32000: Connection closed” 错误,请在 Cherry Studio 项目的该issue界面寻找解决方法:https://github.com/CherryHQ/cherry-studio/issues/3264
方法二:本地安装(高级)
需要 Python 3.10+ 和 uv (推荐) 或 pip。 此方法适合开发者或需要手动配置环境的用户。
使用 uv:
# 克隆仓库
git clone https://github.com/your-username/UniArticles_MCPserver.git
cd UniArticles_MCPserver
# 同步依赖并运行
uv sync
uv run uniarticles-mcp使用 pip:
# 克隆并设置虚拟环境
python -m venv .venv
.venv\Scripts\activate
# 安装依赖
pip install -e .
# 运行
python -m uniarticles配置说明
在项目根目录创建 .env 文件:
ELSEVIER_API_KEY=your_elsevier_api_key
# 必须配置。可以在 https://dev.elsevier.com/ 中申请。
NCBI_API_KEY=your_ncbi_api_key
# 可选。NCBI Entrez 无此 Key 也可用;配置后仅将 PubMed 限速从 3 请求/秒
# 提升到 10 请求/秒。免费申请:先登录 https://www.ncbi.nlm.nih.gov/
# 再前往 https://account.ncbi.nlm.nih.gov/settings/
# 可选。CORE 无此 Key 也可用,但额度低得多(未认证档 100 tokens/天、
# 10 次/分钟,且不提供 fullText;配置后 1,000 tokens/天、25 次/分钟)。
# 使用 CORE 系工具时强烈建议配置。免费申请:https://core.ac.uk/services/api#form
CORE_API_KEY=your_core_api_key项目结构
src/
└── uniarticles/
├── server.py # MCP Server 入口点
└── sources/ # 数据源模块
├── arxiv.py
├── pubmed.py
├── scopus.py
└── ...
pyproject.toml # 项目元数据与依赖验证安装
本项目未附带独立的测试套件;请通过启动服务来验证安装是否成功。服务通过 stdio 通信,启动成功后会保持运行并静默等待客户端发来的 JSON-RPC 输入(按 Ctrl+C 退出):
uv run uniarticles-mcp # 使用 uv 安装时
# 或
python -m uniarticles # 使用 pip 安装时若进程启动过程中没有出现导入或配置错误,即表示安装正常。
可用工具列表
以下工具按数据源分组,每个数据源一张表格。共注册 29 个工具,只要对应数据源的 Key(如有要求)已配置即可全部使用。每个工具都返回相同的归一化 JSON 结构(ok、source、query、count、items、error)。
Scopus
工具名 | 参数 | 说明 |
|
| 按查询串搜索 Scopus 文档。默认按相关度排序,因此用标题检索能直接返回目标文献本身,而不是最新的松散匹配;如需按日期排序请显式传 |
|
| 按 EID 获取归一化的摘要记录(标题、作者、机构、期刊、标识符)。摘要正文仅在更高级别、受订阅限制的视图下才会返回。 |
|
| 按 ISSN 查询期刊/连续出版物元数据(出版商、Open Access 状态、收录年份、学科领域、期刊主页)。 |
| (无) | 检查 Elsevier API 用量/速率限制状态(通过 Scopus 端点)。 |
|
| 按期刊名、出版商、学科、Open Access 状态等多个可选条件搜索期刊/连续出版物(无需 ISSN),结果含 SNIP/SJR 计量指标。 |
|
| 查询 Scopus/ScienceDirect 学科分类代码,用于构造更精确的检索查询。 |
ScienceDirect
工具名 | 参数 | 说明 |
|
| 按标识符(pii/doi/pubmed_id/eid)检索归一化的文章记录(标题、作者、期刊、标识符、主题)。 |
|
| 获取某篇文章的配图/表格/补充材料的元信息(文件名、MIME 类型、对象类型、下载链接)。仅返回对象清单与链接,不下载二进制内容本身。 |
ArXiv
工具名 | 参数 | 说明 |
|
| 按查询串搜索 arXiv 论文。 |
|
| 列出指定 arXiv 分类下最新提交的论文。 |
|
| 按 ID 获取指定 arXiv 论文的元数据。 |
可用性提示:上游主机 export.arxiv.org 可能偶发卡住。在 2026-09-18 的全量回归中,arxiv_paper_detail_by_id 与 arxiv_latest_paper_list_by_category 均挂起约 5 分钟后以连接超时失败,而同一时刻、同一主机上的 arxiv_paper_search_by_query 在 1.4 秒内正常返回;约 15 分钟后故障自行恢复,三个工具全部正常。那次挂起原本是无上限的:arxiv.Client 根本不暴露任何超时参数,只有 page_size / delay_seconds / num_retries。自 v3.4.0 起,客户端改为单次请求 15 秒超时 + 整体 45 秒兜底,因此上游卡住时会快速失败并给出同时标明两个上限的可操作错误,而不是长时间挂起。_verify/arxiv_timeout_check.py 通过把客户端指向一个"只接受连接、从不响应"的本机监听来离线复现该行为;若真的遇到超时,可运行 _verify/arxiv_connectivity_test.py 做 DNS→TCP→TLS→HTTP 分层诊断,以区分上游卡顿与本机网络问题。
PubMed(NCBI Entrez)
以下工具直连 NCBI E-utilities。无 Key 即可使用;配置可选的 NCBI_API_KEY(NCBI 免费申请)仅将限速从 3 请求/秒提升到 10 请求/秒。
工具名 | 参数 | 说明 |
|
| 按关键词检索(ESearch + EFetch),返回归一化记录(标题、摘要、作者、期刊、doi、pmid、pmcid、关键词、日期)。 |
|
| 对一批 PMID 做轻量元数据批量查询(ESummary),含检索工具没有的字段(pmcid、pubstatus、pmcrefcount、elocationid)。无效 PMID 会作为带 |
|
| 查询与某 PMID 主题相关的 PubMed 文献(ELink“相似文献”),返回相关 PMID 列表(已剔除该 PMID 自身)。 |
|
| 查询某 PMID 的 PubMed Central 关联—— |
Crossref
工具名 | 参数 | 说明 |
|
| 按关键词检索文献。无需 Key。 |
|
| 按 DOI 查询单篇文献。无需 Key。 |
Europe PMC
工具名 | 参数 | 说明 |
|
| 检索 Europe PMC(EBI 生命科学聚合库,区别于 NCBI PubMed),仅返回首页结果。无需 Key。 |
DOAJ
工具名 | 参数 | 说明 |
|
| 检索开放获取期刊目录(DOAJ)。无需 Key。 |
OpenAIRE
工具名 | 参数 | 说明 |
|
| 检索 OpenAIRE(欧洲开放科学聚合库)。无需 Key。 |
CORE
CORE 在本版本由 1 个工具扩展为 9 个(其余 8 个数据源共 20 个工具,合计 29)。它是本服务器唯一同时提供"分布统计"(年 / 出版社 / 学科等 facet)与"机构库画像"(某篇论文被哪些机构库采集)的数据源,其余 8 个源都只有"检索列表"或"按标识符取单条"。
work 与 output 的区别(选工具前先看这一句):work 是 CORE 去重后的作品级记录(同一篇论文只有一条);output 是未经去重的原始采集信号(同一篇论文在几个机构库被采集就有几条)。要论文本身用 works 系工具,要看某个采集副本的许可/仓库信息才用 outputs 系工具。
工具名 | 参数 | 说明 |
|
| 按关键词检索 CORE 作品。 |
|
| 按裸 DOI(如 |
|
| 取某作品在各机构库中的版本实例列表(未去重的采集副本),各项含 |
|
| 取作品的生命周期时间戳( |
|
| 按关键词统计 CORE 文献的分布(年 / 作者 / 出版社 / 学科等)。返回的不是文献列表,而是每个维度一项( |
|
| 按关键词检索 CORE 的机构库 / 期刊源(data providers,不是论文)。返回 |
|
| 按数字 ID 取机构库详情。两个可选开关各追加一次上游请求: |
|
| 按数字 ID 取原始采集记录详情( |
|
| 按关键词检索原始采集记录(outputs,未去重)。建议使用 |
全部 CORE 工具均无条件注册(无 Key 也能调用,只是额度更低)。触发限流时错误信息会给出可重试时间与当前额度,并提示配置
CORE_API_KEY。
可参考agent提示词
你是一名文献检索助手,已接入 UniArticles 的 MCP 工具。
请严格按以下四步依次执行,不要跳步。
第一步 —— 检索之前,先拆分我的需求。
用一句话复述我到底要什么,并拆出以下要素:
(a)主题/研究问题;(b)检索类型:广泛扫描 / 定位某一篇已知文献 / 按作者或期刊查找;
(c)学科领域;(d)时间窗口;(e)语言;(f)需要多少篇算够用。
第二步 —— 判断哪些数据源「一定不匹配」,并明确告诉我。
按以下规则直接跳过,不要调用这些源:
- 主题仅属生物医学/生命科学 —— 排除 arXiv(学科不对口)。注意 DOAJ、CORE、
OpenAIRE 是全学科的开放获取聚合源,生物医学一样覆盖,不要按学科排除它们;
它们是否该排除只取决于下一条。
- 主题属物理、数学、计算机、统计、定量生物学、经济学 —— arXiv 可用;
其他学科(人文社科、临床医学等)—— 排除 arXiv(它只有预印本,没有期刊覆盖)。
- 我要找同行评审 / 主流期刊 / 非开放获取的文献 —— 排除 DOAJ、CORE、OpenAIRE
(三者只收开放获取内容,会把结果集带偏)。
- 我要全文或 PDF —— UniArticles 的任何工具都不返回全文或二进制文件。
请先说明这一点,然后只把各源用作「元数据 + 链接」。
- 我要非英文(如中文)文献 —— 本服务器不收录 CNKI / 万方 / 维普,没有任何中文
数据库源。实测「深度学习」这类中文查询:Crossref、DOAJ、CORE 能返回中文题录,
Europe PMC 返回中文期刊的英译题录(标题带方括号),Scopus 命中不稳定(同一查询
0~1 条),PubMed 与 arXiv 为 0 条。请如实说明覆盖率远低于中文数据库,不要声称
可以替代 CNKI/万方。
- Scopus、ScienceDirect —— 仅在 `ELSEVIER_API_KEY` 已配置时可用。若调用返回授权/
配额错误,把该源标记为「不可用」后继续,不要悄悄放弃这部分需求。
无论如何至少保留两个数据源。
第三步 —— 在剩下的数据源中按以下顺序检索,并遵循对应的查询写法。
1. Scopus —— 定位已知文献用 TITLE("完整标题");主题检索用
TITLE-ABS-KEY(词 AND 词);count 取 5–10。
2. Crossref —— 普通关键词检索;同时用它核对每篇文献的 DOI。
3. PubMed(仅生物医学)—— 定位已知文献用 完整标题[Title],且不要加引号,
加了引号反而返回 0 条。不要直接传一长句自然语言:诸如 "in" 这类停用词
会让整条查询归零。多个词组之间请显式使用 AND。
4. Europe PMC —— 字段语法与 PubMed 一致,例如 TITLE:"完整标题"。
5. arXiv(仅限预印本学科)—— 定位已知文献用 ti:"完整标题";主题检索用 all:词。
6. DOAJ、CORE、OpenAIRE —— 仅开放获取。DOAJ 的相关度排序偏弱,
用标题式查询会返回明显离题的结果,因此每一条都必须先核对标题再写进结果。
CORE 除了检索,还能给出**分布统计**(core_work_aggregate_by_query,比如这批
文献都发在哪些年/出版社)与**机构库画像**(core_data_provider_* /
core_work_outputs_by_id,比如某篇论文被哪些机构库采集);需要这类信息时
直接调用对应工具,不要用其它源的结果自行拼凑。
7. ScienceDirect —— 只能按标识符查询(DOI/PII),它没有检索工具,不要试图检索。
每个源取 5–10 条。同一主题在多个源各查一遍是预期用法,而不是「失败后降级」。
任何源报错就跳过,并记录下来。
第四步 —— 汇总成表并汇报。
先按 DOI 去重,再按归一化标题去重。只输出一张 Markdown 表格,按发表时间由新到旧:
| 文献标题 | 标题翻译 | 发表时间 | 期刊/会议 | DOI 链接 | 文献源 | 内容介绍 |
- 标题翻译:把文献标题翻译成中文;若原标题已是中文,此列填「—」。
- DOI 链接:[10.xxxx/yyy](https://doi.org/10.xxxx/yyy) 格式;若无 DOI,则给出该源的原始链接。
- 内容介绍:1–2 句,且只能依据工具真实返回的摘要撰写。
没有摘要时写「无摘要」,严禁自行编造或推测内容。
- 文献源:填工具返回的 `source` 值;同一篇文献被多个源命中时全部列出。
表格之后,再列出:跳过了哪些源及原因,以及哪些查询返回了 0 条结果。
我的需求:<在这里写下你要查找的内容>🤝 贡献与共建
囿于笔者主修化学方向,对其他研究方向的数据库及API开发情况不甚了解,欢迎有志之士提出PR、贡献其他数据源。
⚖️ 协议与致谢
协议
双许可:AGPL-3.0-or-later 或 商业授权
本项目以 GNU Affero 通用公共许可证 v3.0 或更高版本(AGPL-3.0-or-later) 开源发布,全文见 LICENSE。你可以据此自由使用、修改与再分发,包括商业用途——前提是遵守 AGPL 条款:若你分发修改后的版本,或将其作为网络服务提供给用户,必须向这些用户提供对应的源代码。
💼 商业授权: 若 AGPL 的传染性条款不适合你的场景——例如需要闭源集成到商业产品中,或需要在不公开修改的前提下将修改版作为网络服务运营——可另行联系作者获取商业授权:wangzh685@mail2.sysu.edu.cn。
说明:此前“AGPL-3.0 with commercial restriction(商业使用受限)”的表述不准确,已在 v3.4.0 更正。AGPL 并不限制商业使用,它限制的是闭源再分发与闭源网络服务。
特别致谢
ScopusMCP: ScopusMCP是笔者第一个开发成功的文献检索MCP工具,但初始相当臃肿与难以移植,感谢舍友 (https://github.com/qwe4559999) 提供的使用pypi和uv打包的建议。
特别声明
本项目使用了人工智能生成内容。
Available Tools
29 toolsarxiv_latest_paper_list_by_categoryA
List the most recently submitted arXiv papers in a given category
(e.g. 'cs.AI'). Multiple categories may be comma-separated (e.g.
'cs.AI,cs.LG'). Uses arXiv's official cat: query syntax.
Inputs: category (required), max_results.
arxiv_paper_detail_by_idC
Get detailed information (abstract/metadata) for a specific ArXiv paper.
Inputs: paper_id (required).
arxiv_paper_search_by_queryA
Search for papers in ArXiv using a query string.
query is passed to arXiv verbatim, so arXiv's field prefixes work —
use them to pin down one known paper, e.g.
ti:"Attention Is All You Need", au:Vaswani,
abs:transformer, cat:cs.LG. A bare title string is a loose
full-text query and will not reliably return the paper itself.
Inputs: query (required), max_results.
core_data_provider_detail_by_idA
Fetch one CORE data provider (repository / journal source) by numeric id.
include_stats and include_outputs each add one extra upstream request, so
they are off by default; turn them on only when you need the figures. Get the
numeric id from core_data_provider_search_by_query or from the
data_providers field of a work record.
The two sub-resources are best-effort: if the main detail call succeeds but a
sub-resource fails, the response stays ok=True and the failure is reported
under that sub-key (stats / outputs holding an {ok, error} object)
instead of discarding the detail you already have.
Inputs: provider_id (required), include_stats, include_outputs.
core_data_provider_search_by_queryA
Search CORE's data providers — institutional repositories and journal sources.
Each item is a repository/journal source (id / name / type / url /
software / country_code …), not a paper. Take an id from here into
core_data_provider_detail_by_id, or reach a provider id from a work by
reading data_providers in core_work_detail_by_identifier / search results.
max_results is capped at 200 (verified against the real API in buildlog
step 69's F12 plus this step's limit probe: 50/100/200 all accepted).
Inputs: query (required), max_results.
core_output_detail_by_idA
Fetch one raw CORE harvesting record (output) by numeric id.
An output is a de-duplicated source signal: the same paper shows up once
per repository that harvested it, whereas a work is CORE's merged record
for that paper. Call this when you need per-repository detail — license,
repositories, sdg, fulltext_status, source_fulltext_urls — and use
the works tools (core_work_detail_by_identifier) for the paper itself.
output_id must be numeric. Returns a single item.
Inputs: output_id (required).
core_output_search_by_queryA
Search CORE's raw harvesting records (outputs, not de-duplicated).
Difference from core_work_search_by_query: a work is CORE's merged
record for a paper (one per paper), while an output is the per-repository
signal CORE harvested (the same paper can appear several times, once per
repository). Use this tool to find a specific repository's copy of a paper
and inspect its license / fulltext_status / repositories; use the works
tools for the paper itself.
query accepts CORE's own syntax. Field-qualified forms such as
title:"..." and doi:"..." are recommended: this endpoint has
historically returned HTTP 500 (an upstream Azure Search expression error)
for some query expressions — an upstream behaviour, not something this
server retries around, so a 500 surfaces as-is with a hint to try a
field-qualified form.
max_results is capped at 100 and the upstream response inlines fullText
for every hit; the returned items keep metadata and links only.
Inputs: query (required), max_results.
core_work_aggregate_by_queryA
Summarise the distribution of CORE works matching a keyword query.
This is a facet/distribution view, NOT a result list: each item is one
dimension (field / total_buckets / top[{value, count}]), with top
sorted by count descending and truncated to top_n (default 10, capped at
50). count in the response is therefore the number of dimensions
returned, not the number of papers.
query uses the same syntax as core_work_search_by_query. fields is
optional: leave it unset to let CORE pick its default dimensions, or pass
explicit camelCase dimension names (e.g. ["yearPublished", "authors", "publisher"]). Dimension names are passed through verbatim — CORE's
default set is snake_case (year_published, field_of_study) while an
explicit request is camelCase, and both are returned as-is. Each dimension
is capped by upstream at 100 buckets, so total_buckets is also capped at
100 and does not reveal the untruncated cardinality.
Aggregation queries cost more tokens than a plain search upstream; call it when you actually need a distribution, not after every search.
Inputs: query (required), top_n, fields.
core_work_detail_by_identifierA
Fetch the full CORE record of one work by bare DOI or numeric CORE ID.
Use the bare DOI form (e.g. 10.1038/nature12373) — the doi: prefix is
not a valid path segment upstream and returns 404. Both identifier forms
were verified against the real API in buildlog step 69 (F7).
The detail record carries dataProviders / outputs / identifiers that
keyword search results do not include. outputs here are URLs only; call
core_work_outputs_by_id to expand them. Returns a single item.
Inputs: identifier (required).
core_work_outputs_by_idA
List the per-repository copies (outputs) CORE harvested for one work.
identifier MUST be a numeric CORE ID (e.g. 171513974); this sub-resource
rejects DOIs with a 404 (verified in buildlog step 69's F8/F11). If you only
have a DOI, call core_work_detail_by_identifier first and take the numeric
id from its core_id / identifiers fields.
Each item is a de-duplication source record, not the paper text: it carries
download_url / license / fulltext_status / data_provider, which is how
you tell how many repository copies exist and under what licence.
Inputs: identifier (required).
core_work_search_by_queryA
Search CORE (global open-access aggregator) by keyword, with paging.
query accepts CORE's own query syntax (field qualifiers such as
title:"..." / doi:"...", boolean operators, phrase matching).
max_results is capped at 100 (CORE's per-request ceiling) and offset
pages through the result set; note that a large max_results returns a big
payload (roughly 435 KB for 100 hits) and can take up to ~45 s on slow
networks, so keep the default 10 unless you need more.
Returns metadata and download links only — the upstream fullText field is
excluded at the request level and never surfaces. Works without an API key,
but the anonymous tier is token-metered (100 tokens/day, 10 requests per
minute, no fullText); configuring CORE_API_KEY raises it to 1,000 tokens/day
at 25 requests per minute.
Inputs: query (required), offset, max_results.
core_work_stats_by_idA
Fetch the lifecycle timestamps of one CORE work (deposited / published / updated / accepted).
identifier accepts either a bare DOI or a numeric CORE ID — unlike
core_work_outputs_by_id, this endpoint does not require a numeric id
(both verified in buildlog step 69's F9/F9b). Returns a single item with
only those timestamp fields, because that is all the upstream returns.
Inputs: identifier (required).
crossref_work_detail_by_doiA
Look up a single work on Crossref by DOI. No API key needed.
Inputs: doi (required).
crossref_work_search_by_queryB
Search Crossref (DOI registration agency metadata) for works by keyword. No API key needed.
Inputs: query (required), max_results.
doaj_article_search_by_queryA
Search DOAJ (Directory of Open Access Journals) for open-access articles by keyword. No API key needed.
Inputs: query (required), max_results.
europepmc_paper_search_by_queryA
Search Europe PMC (EBI's life-sciences literature aggregator, distinct from NCBI PubMed) for papers by keyword. Returns the first page of results only (result ordering is Europe PMC's relevance ranking). No API key needed.
Inputs: query (required), max_results.
openaire_research_product_search_by_queryB
Search OpenAIRE (European open science aggregator) for research products by keyword. No API key needed. NOTE: reference projects reported occasional HTTP 403 from this service; it was not reproduced during this project's probing. A network error here may reflect that known intermittency, not a tool bug.
Inputs: query (required), max_results.
pubmed_paper_search_by_queryA
Search PubMed by keyword via NCBI Entrez (ESearch to get PMIDs, then EFetch to retrieve and parse the article XML). Returns normalized records (title, abstract, authors, journal, doi, pmid, pmcid, keywords, date).
query is passed to NCBI verbatim, so PubMed field tags and MeSH
terms work. To pin down one known paper, use the title tag WITHOUT
quotes, e.g. Genome engineering using the CRISPR-Cas9 system[Title]
— wrapping the title in quotes together with the tag
("..."[Title]) makes NCBI return 0 results.
Inputs: query (required), max_results.
pubmed_paper_summary_lookup_by_pmidsA
Look up lightweight PubMed metadata for a batch of PMIDs via ESummary.
Faster than a full fetch and carries fields the search tool lacks (pmcid,
pubstatus, pmcrefcount, elocationid). Invalid PMIDs are returned as items
with a per-item error field rather than failing the whole call. Max 200
PMIDs per call (excess is truncated).
Inputs: pmids (required).
pubmed_pmc_linkage_lookup_by_pmidA
Look up a PMID's PubMed Central (PMC) linkages. Returns TWO distinct groups, kept separate on purpose: 'own_pmc_fulltext' (this article's own open-access PMC full-text record, if any — check 'has_pmc_fulltext') and 'cited_by_pmc_articles' (other PMC articles that cite it). Both empty is a normal result, not an error.
Inputs: pmid (required).
sciencedirect_article_object_by_identifierA
Get metadata (filename, mimetype, type, download link) for figures/tables/ supplementary materials attached to an article. Does NOT download the binary content itself -- only returns the metadata list and download links. identifier_type: doi or pii (both verified working); scopus_id/pubmed_id may also work.
Inputs: identifier (required), view, identifier_type.
sciencedirect_article_retrieve_by_identifierA
Retrieve an article record by identifier type (pii, doi, pubmed_id, eid) and value. Returns a normalized record (title, authors, journal, identifiers, subjects, etc.). Default view is META (unrestricted); the full-text body is only populated under richer, entitlement-gated views.
Inputs: identifier (required), view, identifier_type.
scopus_abstract_detail_by_eidA
Get detailed abstract information for a Scopus document (by EID). Returns a normalized record (title, authors, affiliations, journal, identifiers, etc.). Default view is META (unrestricted); the abstract body is only populated under richer views such as FULL/META_ABS if your subscription supports them.
Inputs: eid (required), view.
scopus_api_usage_statusA
Check current Elsevier API usage/rate-limit status (via Scopus endpoint).
Inputs: no named parameters.
scopus_document_search_by_queryA
Search for documents in Scopus using a query string.
query is passed to Scopus verbatim, so Scopus's own field syntax is
available — use it when hunting one specific paper, e.g.
TITLE("Attention Is All You Need"), DOI(10.1016/...) or
AUTHKEY(Smith). A bare title string is instead treated as a loose
keyword query, which surfaces related-but-different papers ahead of the
target. sort defaults to relevancy; pass sort="coverDate"
for newest-first ordering.
Inputs: query (required), sort, view, count.
scopus_serial_title_by_issnC
Get journal/serial metadata (title, publisher, Open Access status, coverage years, subject areas, homepage) by ISSN. Default view is STANDARD (verified working with a basic subscription tier).
Inputs: issn (required), view.
scopus_serial_title_search_by_criteriaA
Search journals/serials by multiple optional criteria (no ISSN required). Sibling of scopus_serial_title_by_issn, which looks up ONE journal by exact ISSN; this tool searches across journals and returns a list, including SNIP/SJR metrics. Optional filters: title (substring match), issn, pub (publisher), subj (subject ABBREVIATION such as COMP/CHEM, NOT the numeric code), content (journal/tradejournal/conferenceproceeding/bookseries), date (year), oa (all/full/partial/none), start (page offset), count (page size, max 200). view defaults to STANDARD (ENHANCED/CITESCORE may require a higher subscription). Passing no filter is allowed but browses ALL serials (large) — supply at least one.
Inputs: oa, pub, date, issn, +6 more.
scopus_subject_classification_lookup_by_sourceA
Look up Scopus/ScienceDirect subject classification codes (to help build
more precise search queries). source is required and must be 'scopus' or
'scidir'. Optional filters: description, detail, code, abbrev, field (field
restricts which fields are returned). scidir results additionally carry
parent_code (hierarchy); scopus results have parent_code = null.
Inputs: source (required), code, field, abbrev, +2 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 29 tools
The tools are largely distinct by their target source (Scopus, PubMed, arXiv, etc.). However, there is some potential confusion between tools like 'core_work_search_by_query' and 'core_output_search_by_query' or 'pubmed_paper_search_by_query' vs 'europepmc_paper_search_by_query' if an agent doesn't read carefully, but descriptions clarify the differences.
Most tool names follow a clear pattern of '<source>_<action>_<object>_by_<identifier>' (e.g., scopus_document_search_by_query, arxiv_paper_detail_by_id). Deviations exist like 'arxiv_latest_paper_list_by_category' and 'scopus_api_usage_status' which break the pattern slightly, but overall it is consistent and readable.
With 29 tools, the count is on the high side but not extreme. The server aggregates multiple scholarly APIs (Scopus, arXiv, PubMed, Crossref, CORE, etc.), so many tools are needed. However, it feels slightly heavy and could be streamlined (e.g., combining some CORE output tools).
The domain is scholarly literature search and retrieval across multiple sources. Each source has at least a search and detail tool, covering core workflows. Minor gaps: no update or delete operations (not applicable for read-only APIs), but missing citation or full-text retrieval for some sources (e.g., no direct arXiv full-text download, no PubMed Central full-text retrieval beyond linkage).
Maintenance
Related MCP Connectors
arXiv and OpenAlex/Crossref paper search (an open Google Scholar alternative) as agent tools.
Academic literature search, retrieval, and private library management on top of OpenAlex.
Scholarly search: OpenAlex, Crossref, arXiv, OpenCitations and PubMed in one endpoint.
Federated search of books and papers, BibTeX/RIS citations, open-access retrieval and reading.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables LLMs to search, analyze, and summarize academic research papers in real-time from arXiv, Semantic Scholar, and PubMed. Provides automatic deduplication, citation analysis, and BibTeX generation across multiple research databases.22 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables users to search and analyze academic papers from multiple sources, fetch metadata and full text, and build structured outputs like literature maps and paper comparisons.77 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables LLM agents to search academic literature across multiple sources (IEEE, arXiv, ACM, Semantic Scholar, CORE, Scite, Consensus) and index/search local PDFs.MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to conduct literature reviews by screening multiple academic databases, performing backward and forward citation snowballing, and resolving or downloading full-text PDFs through subscription and open-access sources.MIT