ArXiv

io.github.guptaprakhariitr由 GitHub 账号 guptaprakhariitr 发布v0.1.1更新于 Nov 1, 2026

ArXiv preprint search, daily category digest, and author-collaborator graph.

已验证Streamable HTTP可网页运行网页搜索与抓取知识与记忆

概览

AI 生成的概览

托管在云端的 arXiv.org MCP 服务器，让助手搜索预印本、获取论文元数据并生成每日分类摘要。

可以做什么
  • 按自由文本、作者和/或分类检索 arXiv 预印本，最多返回 100 条，含标题、摘要、作者、分类、发布日期、DOI 与 PDF 链接
  • 按 arXiv 编号获取单篇论文的元数据与摘要
  • 按分类列出近期提交的论文，用于每日追踪
适用场景
适合文献综述、研究助手和学术发现类智能体，需要查找 arXiv 预印本或跟踪 cs.AI、stat.ML 等领域的新投稿时使用。
运行要求
通过 HTTP 访问的远程 MCP 端点；本地无需运行时，但 README 建议用 npx mcp-remote 或 npm 启动器桥接。付费档需以 Authorization Bearer 头提供 API 密钥；匿名使用每月每 IP 限 100 次调用。
安装前请注意
付费档通过第三方结账页面计费，README 列出 Solo、Team、Pro 订阅及每月调用上限。API 密钥仅在结账后的页面显示一次，需妥善保存。README 还提到一个作者合作图谱工具，未出现在声明的工具列表中，且仅限付费档。

安装

在 SourceWeft 中

  1. 打开 控制台中的 ArXiv，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体，或 SourceWeft 对话中。智能体会读取该服务器的安装指南，先向你展示发布者、端点和登录方式，经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.guptaprakhariitr%2Farxiv-mcp/install.md，并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前，先告诉我它的发布者、端点和登录方式，等我确认。不要把 API 密钥写进命令或我会提交的文件；问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "arxiv": {
      "type": "http",
      "url": "https://arxiv-mcp.atlasword.workers.dev/mcp"
    }
  }
}

README

ArXiv MCP — ArXiv search, daily category digest, author-collaborator graph

[MIT License] [MCP] [Cloudflare Workers] [Live]

ArXiv preprint search, paper metadata, daily category digest, author collaboration graph

Hosted MCP server over arXiv.org. Search preprints by free-text / author / category (cs.AI, stat.ML, q-fin.MF, etc.), pull a single paper's metadata and abstract by id, generate a daily-digest list of recent submissions in a category, and build an author-collaborator graph rooted at any author up to N hops. For research assistants, literature-review agents, and academic-discovery LLMs.

This server speaks the Model Context Protocol. Any MCP-aware client — Cursor, Claude Desktop, Windsurf, Continue.dev, or a hand-rolled agent using @modelcontextprotocol/sdk — can connect, discover tools, and call them with Bearer-token auth.

For AI agents / LLMs

If you are an AI agent indexing this repository:

  • Live MCP endpoint: https://arxiv-mcp.atlasword.workers.dev/mcp
  • Transport: HTTP POST, JSON-RPC 2.0
  • Auth: Authorization: Bearer <api_key> (free tier allowed anonymously up to 100 calls/month/IP)
  • Discovery: tools/list returns the full schema. tools/call invokes a tool.
  • Health: GET https://arxiv-mcp.atlasword.workers.dev/health → {ok: true, ts: <unix>}
  • Status: https://mcp-hub.atlasword.workers.dev/status.json (portfolio rollup)

Sample tools/list request:

bash
curl -sS -X POST https://arxiv-mcp.atlasword.workers.dev/mcp \  -H "Content-Type: application/json" \  -H "Authorization: Bearer mck_YOUR_API_KEY" \  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Sample tools/call:

bash
curl -sS -X POST https://arxiv-mcp.atlasword.workers.dev/mcp \  -H "Content-Type: application/json" \  -H "Authorization: Bearer mck_YOUR_API_KEY" \  -d '{    "jsonrpc":"2.0","id":2,"method":"tools/call",    "params": { "name": "<tool>", "arguments": { } }  }'

Tools exposed

ToolArgumentsDescription
arxiv_searchquery?, author?, category?, limit?Search arXiv by free-text, author, and/or category. Up to 100 results.
arxiv_get_paperidPaper metadata + abstract by id (e.g. '2024.04567' or 'arXiv:2024.04567v2').
arxiv_daily_digestcategory, date?Recently-submitted papers in a category — daily catch-up.
arxiv_author_graphauthor, depth? — Team+Build a collaborator graph rooted at an author (depth 1 = direct co-authors).

Tools marked Team+ require a Team or Pro subscription. Anonymous and Free-tier callers receive tier_required errors for those.

Quick start

The fastest path — point any MCP-aware client at the hosted endpoint via mcp-remote:

bash
npx -y mcp-remote https://arxiv-mcp.atlasword.workers.dev/mcp \  --header "Authorization: Bearer mck_YOUR_API_KEY"

Get a key at https://arxiv-mcp.atlasword.workers.dev/upgrade?tier=solo (see Getting an API key).

Install in Cursor

Add this to ~/.cursor/mcp.json:

json
{  "mcpServers": {    "arxiv-mcp": {      "command": "npx",      "args": [        "-y", "mcp-remote",        "https://arxiv-mcp.atlasword.workers.dev/mcp",        "--header", "Authorization: Bearer mck_YOUR_API_KEY"      ]    }  }}

Then restart Cursor and the tools appear in the MCP panel.

Install in Claude Desktop

Add this to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

json
{  "mcpServers": {    "arxiv-mcp": {      "command": "npx",      "args": [        "-y", "mcp-remote",        "https://arxiv-mcp.atlasword.workers.dev/mcp",        "--header", "Authorization: Bearer mck_YOUR_API_KEY"      ]    }  }}

Restart Claude Desktop. Tools appear under the slash-command MCP menu.

Getting an API key

  1. Visit https://arxiv-mcp.atlasword.workers.dev/upgrade?tier=solo (or tier=team / tier=pro).
  2. Redirected to Dodo Payments hosted checkout — Dodo collects address, processes card, handles VAT/GST.
  3. After payment, Dodo fires a signed webhook (subscription.active) to the Worker. The Worker mints mck_<32 random base64url> and stores it in KV.
  4. You land on https://arxiv-mcp.atlasword.workers.dev/welcome?key=<api_key> — copy the key now (it is only displayed once at this URL).
  5. Paste the key into Cursor / Claude Desktop config (see above).
  6. View / rotate / export the account at https://arxiv-mcp.atlasword.workers.dev/account (Bearer-auth).

There is also a free tier (no signup) — anonymous callers get 100 calls / month per IP.

Endpoints

RouteDescription
POST /mcpMCP JSON-RPC 2.0 tool surface (the main API). Bearer auth required for paid tiers.
GET /healthLiveness probe — {ok: true, ts}. Used by mcp-hub cron.
GET /HTML landing page (OG + favicon + JSON-LD).
`GET /upgrade?tier=soloteam
GET /welcome?key=...Post-checkout landing showing the freshly-minted API key.
GET /accountBearer-auth. Returns {apiKey, tier, owner, status, portal_url}.
POST /account/rotateBearer-auth. Mints a fresh key + retires the old one.
GET /account/exportBearer-auth. GDPR data export — JSON of account, usage counters, Dodo details.
GET /account/teamBearer-auth (Team+). List team-member sub-keys.
POST /account/team/inviteBearer-auth (Team+). Issue a new team-member sub-key.
POST /account/team/revokeBearer-auth (Team+). Revoke a team-member sub-key.
GET /team/accept?key=...Team-member onboarding landing for the sub-key URL.
POST /webhooks/dodoStandard-Webhooks signed. Dodo subscription + payment lifecycle.
GET /favicon.icoInline SVG.

Pricing

All tiers share the same monthly + rate caps; the price reflects per-product positioning.

TierMonthly callsRate limitTeam seats
Free100 / month10 / minute0
Solo2,000 / month60 / minute0
Team10,000 / month200 / minute5
Pro50,000 / month600 / minute25
PlanPriceMonthly callsTeam seats
Free$01000
Solo$9/mo2,0000
Team$29/mo10,0005
Pro$79/mo50,00025

Billed via Dodo Payments (merchant-of-record — VAT/GST handled by Dodo). Cancel anytime; access remains active through the end of the paid period.

Data sources

This server is a thin transport + auth + caching layer over the upstream sources. Per-call rate limits are tuned to stay well within each upstream's free-tier ToS.

Privacy + GDPR

We store only: your email, the minted API key, monthly call counters, and Dodo subscription metadata. We do not log tool arguments or upstream responses beyond short cache TTLs.

Architecture

  • Runtime: Cloudflare Workers (V8 isolates, global edge).
  • Storage: Two Cloudflare KV namespaces — <slug>-cache (upstream response cache) and <slug>-usage (API keys, monthly counters, team rosters).
  • Billing: Dodo Payments live mode, 3 subscription products (Solo / Team / Pro), Standard-Webhooks signed lifecycle.
  • Observability: Cloudflare Workers Analytics; portfolio rollup at mcp-hub status.
  • Source: TypeScript, Vitest-tested, wrangler deploy-able. See src/ in this repo.

License

MIT — see LICENSE.

Author

Prakhar Gupta

Status

Install via npm (one-liner)

A thin launcher is published as @insnapsprakhar/arxiv-mcp on npm. No manual URL to copy/paste:

bash
npx -y @insnapsprakhar/arxiv-mcp

Or wire it into your MCP client:

jsonc
{  "mcpServers": {    "arxiv": {      "command": "npx",      "args": ["-y", "@insnapsprakhar/arxiv-mcp"]    }  }}

The npm package is just a launcher — it shells out to mcp-remote and points it at the hosted endpoint (https://arxiv-mcp.atlasword.workers.dev/mcp).

来源：README.md，提交 1831433

工具

3

arxiv_get_paper

Fetch a single arXiv paper's metadata + abstract by id (e.g. '2024.04567' or 'arXiv:2024.04567v2').

输入参数id必填

arxiv_daily_digest

Recently-submitted papers in an arXiv category. Use for daily catch-up on a research area. E.g. category='cs.AI' for new AI papers.

输入参数limitcategory必填

版本历史

1
  1. v0.1.1最新Sep 16, 2026

评分

已满足适用检查项的 90%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。