The scipdf-mcp server provides MCP tools for academic paper PDF downloading, management, and citation processing.
Download papers: Single papers by DOI, title, URL, or citation; batch download with BibTeX/RIS support, deduplication, and manifest output. Supports Sci-Hub, Unpaywall, and publisher APIs.
Resolve DOIs: Resolve titles, URLs, or citations to DOIs via Crossref and OpenAlex.
Parse references: Extract DOIs and queries from BibTeX, RIS, or plain-text reference lists.
Manage mirrors: List configured Sci-Hub mirrors and PDF hosts; check mirror availability with cache bypass.
Unpaywall integration: Lookup OA status and PDF URL for a DOI (requires email).
File operations: List downloaded PDFs by date; open a paper with the system default viewer.
Format citations: Generate APA, GB/T 7714, or BibTeX citations from a DOI.
Extract text: Extract text from downloaded PDFs.
Configuration: Hot-reload server config from environment variables and config.json without restarting.
Enables downloading academic papers by resolving DOIs via Crossref and OpenAlex, manages local paper library, and provides citation formatting.
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., "@scipdf-mcp download paper 10.1038/nature12373 and get citation"
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.
一键安装的 学术论文 PDF 下载 工具：MCP Server + Skill + CLI（v0.3）。
用户找论文 → Crossref/OpenAlex/Semantic Scholar/arXiv 检索
用户要全文 → DOI 或 arXiv ID → 下载 → 返回本地 path + 引用合规提示：请自行确认当地法规与机构政策。本工具仅供个人科研学习自用。
仓库： https://github.com/gjjisadog/scipdf-mcp
请按仓库 AGENTS.md 安装 scipdf-mcp：
git clone https://github.com/gjjisadog/scipdf-mcp.git
cd scipdf-mcp && bash install.sh
重启客户端后用 /scipdf 测试。更新：bash install.sh --updategit clone https://github.com/gjjisadog/scipdf-mcp.git
cd scipdf-mcp
bash install.sh
# 更新
bash install.sh --update要求：Node.js ≥ 20。安装后会注册 MCP + Skill，并做 CLI 自检。
Related MCP server: paper-fetch-mcp
Agent： /scipdf 下载：Nanometre-scale thermometry in a living cell
CLI：
node dist/index.js search "wide bandgap inverter"
node dist/index.js search --sources openalex,semanticscholar --year-from 2022 --oa "SiC inverter"
node dist/index.js citations 10.1038/nature12373
node dist/index.js related 10.1038/nature12373
node dist/index.js search --source arxiv "grid-forming inverter"
node dist/index.js download arXiv:2501.01234
node dist/index.js download 10.1038/nature12373
node dist/index.js download --title "Nanometre-scale thermometry in a living cell"
node dist/index.js batch 10.a/b 10.c/d
node dist/index.js resolve "some title"
node dist/index.js parse refs.bib
node dist/index.js audit refs.bib
node dist/index.js extract "/absolute/path/to/paper.pdf"
node dist/index.js list
node dist/index.js check-mirrors无参数启动 = MCP stdio 服务。
默认路径（无需邮箱）：
arXiv ID / URL → arXiv 官方 PDF（直接下载）
DOI：
（可选）出版商授权 PDF API # Elsevier；Springer Nature/IEEE 需全文 endpoint
（可选）免费 OA 数据源 # SCIPDF_PREFER_OA=true；无需邮箱
→ Unpaywall OA # 另设 SCIPDF_UNPAYWALL_EMAIL 时加入
→ pdfHosts 直链 PDF # 如 sci.bban.top
→ scihubMirrors HTML 镜像 # 逐个解析页面中的 PDF 链接健康缓存会跳过近期失败的镜像（SCIPDF_HEALTH_TTL_MS，默认 15 分钟）。可用 check-mirrors / MCP check_mirrors 探测。
主机 | 说明 |
| 优先尝试的 DOI 直链主机 |
DEFAULT_MIRRORS）内置 15 个 HTML 镜像（以 src/config.ts / config.example.json 为准）：
镜像 | 备注 |
| 传统域名 |
| 传统域名 |
| 传统域名 |
| 传统域名（部分网络较慢或拦截） |
| 传统域名 |
| 传统域名 |
| 传统域名（部分网络 DNS 失败） |
| 已实测可下 PDF |
| 已实测可下 PDF |
| 已实测可下 PDF |
| 已实测；可能跳转 usualwant |
| 已实测可下 PDF |
| 已实测；可能跳转 pismin |
| 已实测可下 PDF |
| 已实测可下 PDF |
镜像可用性随地区与封锁变化；列表仅作默认回退，不是可用性保证。
环境变量（逗号 / 分号 / 换行分隔）：
export SCIPDF_MIRRORS="https://sci-hub.vg/,https://www.pismin.com/"
export SCIPDF_PDF_HOSTS="https://sci.bban.top/pdf/"
node dist/index.js check-mirrors
node dist/index.js download --force 10.1038/nature12373或复制 config.example.json → 项目目录 config.json / ~/.config/scipdf-mcp/config.json，编辑 scihubMirrors 与 pdfHosts。
配置优先级：SCIPDF_* 环境变量 > 配置文件 > 内置默认。
Tool | 说明 |
| Crossref + OpenAlex + Semantic Scholar + arXiv 统一检索 |
| 列出检索源及搜索、引用链、推荐、PDF 等能力 |
| 列出 |
| 通过 Semantic Scholar 前向/后向追踪引用 |
| 查找相似论文 |
| 单篇 DOI/标题/arXiv ID；arXiv 直接从官方 PDF 下载 |
| 批量去重 + manifest v2（来源历史、OA 证据、SHA-256） |
| Crossref + OpenAlex |
| bib/ris/粘贴列表抽 DOI |
| 从下载目录内的 PDF 提取正文并写入相邻 |
| 批量核验引用、解析 DOI、输出规范引用与失败项 |
| 配置与健康缓存探测 |
| 已下载列表 |
| 系统默认打开 PDF |
| APA / GB/T / BibTeX |
| 热加载配置 |
Resource: papers://list
Prompt: download_papers_batch
错误码：DOI_NOT_FOUND / AMBIGUOUS_DOI / INVALID_ARXIV_ID /
MIRROR_BLOCKED / ALL_SOURCES_FAILED / PDF_NOT_IN_DB 等。
变量 | 含义 | 默认 |
| 保存目录 |
|
| 可选，启用 Unpaywall 时使用的真实邮箱 | 未设 = 跳过 Unpaywall |
| 可选，提高 Semantic Scholar 检索配额 | 未设置 = 匿名配额 |
| Elsevier Article Retrieval API key（仅从环境读取） | 未设置 = 跳过 |
| 可选 Elsevier 机构/用户授权令牌 | 未设置 |
| Springer Nature key（仅从环境读取） | 未设置 = 跳过 |
| 已获授权的 Springer Nature PDF endpoint 模板，支持 | 未设置 |
| IEEE API key（仅从环境读取） | 未设置 = 跳过 |
| 已获授权的 IEEE PDF endpoint 模板，支持 | 未设置 |
| 为 true 时才在 Sci-Hub 前试 OA |
|
| 是否允许 Sci-Hub（主路径） |
|
|
|
|
| 直连 PDF 主机（逗号分隔） |
|
| HTML 镜像列表（逗号分隔） | 见上表（15 个） |
| 调试日志 | off |
| 镜像健康缓存 | 15min |
默认只走 Sci-Hub，无需任何邮箱。
若要优先合法 OA，启用：
export SCIPDF_PREFER_OA=true这会先尝试 OpenAlex、Europe PMC 和 Semantic Scholar。若还要加入
Unpaywall，再设置
SCIPDF_UNPAYWALL_EMAIL；该服务要求真实邮箱（统计用量，不收费）。
配置 | 行为 |
默认（无邮箱 / 无 PREFER_OA） | 只走 Sci-Hub / pdfHosts |
只设邮箱 | 仍默认 Sci-Hub（可用 |
仅 | 先免费 OA 数据源，失败再 Sci-Hub |
邮箱 + | 在上述 OA 数据源中加入 Unpaywall |
| 仅 OA |
Grok 示例（可选 OA）：
[mcp_servers.scipdf]
command = "node"
args = ["/path/to/scipdf-mcp/dist/index.js"]
env = {
SCIPDF_DOWNLOAD_DIR = "/Users/you/Documents/Papers",
SCIPDF_UNPAYWALL_EMAIL = "you@gmail.com",
SCIPDF_PREFER_OA = "true"
}见 config.example.json。
search_papers 同时查询 Crossref、OpenAlex、Semantic Scholar 和 arXiv，
将不同来源统一为相同字段，按 DOI、arXiv ID 或“标题 + 年份”去重，并使用
Reciprocal Rank Fusion 融合来源排名。支持：
指定一个或多个来源；
限制年份区间；
最低引用量；
仅保留已知开放获取结果；
最多返回 50 条。
MCP 参数示例：
{
"query": "wide bandgap semiconductor inverter",
"sources": ["openalex", "semanticscholar"],
"limit": 20,
"year_from": 2022,
"min_citations": 5,
"open_access_only": true
}搜索结果只用于发现和筛选论文；选定结果后，把 DOI 或 arXiv ID 交给
download_paper / download_papers 下载。arXiv 结果始终包含官方 PDF
地址，不要求开启 SCIPDF_PREFER_OA。
下载层通过统一 PdfSource 调度：
来源 | 标识符 | 行为 |
| arXiv ID | 官方仓储直接 PDF，合法来源 |
| DOI | 配置 |
| DOI | key + 授权 PDF endpoint；官方标准 Full Text API 返回 JATS/XML，不能冒充 PDF |
| DOI | key + 单独获权的全文 endpoint；普通 Metadata API key 不等于全文权限 |
| DOI |
|
| DOI | 现有 pdfHosts/Sci-Hub 回退链 |
缓存侧车使用 identifier 保存 doi:… 或 arxiv:…，并兼容旧版仅含 doi
的侧车文件。
出版商 key/token 不写入 SciPdfConfig，也不会出现在 MCP 配置回显或
manifest。endpoint URL 中的 api_key、apikey、token 等查询参数在
返回和落盘前会被脱敏。
下载结果继续保留兼容字段 ok、code、source，并新增：
status：downloaded、cached、not_found、not_entitled、
rate_limited、invalid_pdf 等；
attempts：按实际顺序记录来源、结果、耗时、HTTP 状态、合法性和访问模式；
oa / oaEvidence：OA 提供者、版本、许可证及 PDF 证据；
sha256：最终保存或缓存 PDF 的 SHA-256。
批量下载写出的 scipdf-manifest.json 使用 manifestVersion: 2，包含汇总、
上述审计字段和自动凭据脱敏。
npm install && npm run build && npm test
npm run install:allCI：GitHub Actions 上 Node 20/22 build + test。
MIT
Probe Sci-Hub mirrors for availability (uses health cache; force_refresh ignores cache). Only configured public mirrors (or allowlisted public https hosts) are probed — private/localhost URLs are rejected.
Inputs: urls, force_refresh.
Download an academic paper PDF by DOI, title, URL, or citation. Default: Sci-Hub/pdfHosts. Optional Unpaywall OA only if SCIPDF_UNPAYWALL_EMAIL is set AND SCIPDF_PREFER_OA=true. Returns path, source (scihub|unpaywall|cache), citations. Use force=true to re-download.
Inputs: query (required), force, outdir, filename, +2 more.
Batch-download papers. Accepts list of DOIs/titles/URLs, or a single bib/ris/pasted bibliography blob. Dedupes DOIs, writes scipdf-manifest.json, returns indexed results.
Inputs: queries (required), force, outdir, query_type, +1 more.
Format APA / GB/T 7714 / BibTeX from DOI (fetches Crossref metadata).
Inputs: doi (required).
List configured Sci-Hub mirrors, PDF hosts, Unpaywall status, and settings (hot-reloads config).
Inputs: no named parameters.
List PDF files already in the download directory (most recent first).
Inputs: limit, outdir.
Query Unpaywall for Open Access status and PDF URL (does not download). Requires SCIPDF_UNPAYWALL_EMAIL.
Inputs: doi (required).
Open a local PDF path with the system default viewer (macOS open / Windows start / xdg-open).
Inputs: path (required).
Extract DOIs/queries from BibTeX, RIS, or pasted reference list text.
Inputs: text (required).
Hot-reload scipdf config from environment variables and config.json without restarting the process.
Inputs: no named parameters.
Resolve title/URL/citation to DOI via Crossref + OpenAlex. Returns candidates when ambiguous (code AMBIGUOUS_DOI).
Inputs: query (required), query_type.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 11 tools
Each tool has a clearly distinct purpose: checking mirrors vs listing them, single vs batch download, citation formatting vs reference parsing, etc. No two tools have overlapping functionality; even similarly named tools like download_paper and download_papers are differentiated by single vs batch behavior.
All tool names follow a consistent verb_noun pattern in snake_case (e.g., download_paper, list_mirrors, resolve_doi). This makes the tool set predictable and easy for an agent to navigate.
With 11 tools, the server is well-scoped for its domain of academic paper management. Each tool serves a necessary function without redundancy or bloat.
The tool surface covers the full lifecycle: mirror probing, downloading (single and batch), local file listing, citation formatting, reference parsing, DOI resolution, Unpaywall lookup, config reload, and file opening. No obvious gaps for the stated purpose.
Academic research MCP server for paper search, citation checks, graphs, and deep research.
Zotero MCP server for Claude and ChatGPT: search, citations, safe writes, PDF passages and pages.
Document-to-Markdown MCP server — convert PDF, Office and HTML into LLM-ready Markdown.
MCP server for Altmetric APIs - track research attention across news, policy, social media, and more
