arxiv-mcp by oimiragieo/agent-studio
npx skills add https://github.com/oimiragieo/agent-studio --skill arxiv-mcp模式:认知/提示驱动 — 无独立实用脚本;通过智能体上下文使用。
此技能使用现有工具访问 arXiv:
立即生效 - 无需 MCP 服务器,无需重启。
arxiv-mcp 返回学术论文。为防止内存耗尽:
为何限制?
arXiv API 可通过 http://export.arxiv.org/api/query 公开访问。
// ✓ 良好:将结果限制在 20 条
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:transformer+attention&max_results=20&sortBy=relevance',
prompt: '从这些结果中提取论文标题、作者、摘要、arXiv ID 和 PDF 链接',
});
// ✓ 良好:使用特定过滤器减少结果集
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:transformer+attention+2025&max_results=20&sortBy=submittedDate',
prompt: '提取关于 transformer attention 的最新论文',
});
// ✗ 不良:旧行为 - 无限制或超过 20 条结果
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:neural+networks',
// 范围太广 - 将获得数百条结果
});
// ✗ 不良:超出内存限制
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:deep+learning&max_results=100',
// 超出限制 - 内存风险
});
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:transformer+attention&max_results=20&sortBy=relevance',
prompt: '从这些结果中提取论文标题、作者、摘要、arXiv ID 和 PDF 链接',
});
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=au:LeCun&max_results=10&sortBy=submittedDate',
prompt: '提取论文标题、作者、摘要和 arXiv ID',
});
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG&max_results=15&sortBy=submittedDate',
prompt: '提取论文标题、作者、摘要、类别和 arXiv ID',
});
WebFetch({
url: 'http://export.arxiv.org/api/query?id_list=2301.07041',
prompt:
'提取完整详情:标题、所有作者、摘要、类别、发布日期、PDF 链接',
});
| 参数 | 描述 | 示例 |
|---|---|---|
search_query | 带字段前缀的搜索词 | all:transformer, au:LeCun, ti:attention |
id_list | 逗号分隔的 arXiv ID | 2301.07041,2302.13971 |
max_results | 结果数量(默认 10,最大 100) | max_results=20 |
start | 分页偏移量 | start=10 |
sortBy | 排序方式:relevance, lastUpdatedDate, submittedDate | sortBy=submittedDate |
sortOrder | ascending 或 descending | sortOrder=descending |
| 前缀 | 字段 | 示例 |
|---|---|---|
all: | 所有字段 | all:machine+learning |
ti: | 标题 | ti:transformer |
au: | 作者 | au:Vaswani |
abs: | 摘要 | abs:attention+mechanism |
cat: | 类别 | cat:cs.LG |
co: | 评论 | co:accepted |
使用 AND、OR、ANDNOT 组合词条:
search_query=ti:transformer+AND+abs:attention
search_query=au:LeCun+OR+au:Bengio
search_query=cat:cs.LG+ANDNOT+ti:survey
pnpm search:code 或 ripgrep 技能(Grep/Glob 作为备用)arxiv-mcp 最适合:
使用 Exa 进行更自然的语言查询,并支持 arXiv 过滤:
mcp__Exa__web_search_exa({
query: 'site:arxiv.org transformer architecture attention mechanism deep learning',
numResults: 10,
});
mcp__Exa__web_search_exa({
query: 'site:arxiv.org large language model scaling laws 2024',
numResults: 15,
});
mcp__Exa__web_search_exa({
query: 'site:arxiv.org author:"Yann LeCun" deep learning',
numResults: 10,
});
| 类别 | 领域 |
|---|---|
| cs.AI | 人工智能 |
| cs.LG | 机器学习 |
| cs.CL | 计算与语言(NLP) |
| cs.CV | 计算机视觉 |
| cs.SE | 软件工程 |
| cs.CR | 密码学与安全 |
| stat.ML | 机器学习(统计学) |
| math.* | 数学(所有子类别) |
| physics.* | 物理学(所有子类别) |
| q-bio.* | 定量生物学 |
| econ.* | 经济学 |
// 从广泛的 Exa 搜索开始,进行语义匹配
mcp__Exa__web_search_exa({
query: 'site:arxiv.org transformer attention mechanism neural networks',
numResults: 10,
});
// 通过 ID 获取感兴趣论文的详细信息
WebFetch({
url: 'http://export.arxiv.org/api/query?id_list=2301.07041,2302.13971',
prompt: '为每篇论文提取完整元数据:标题、作者、摘要、类别、PDF URL',
});
// 按感兴趣论文的类别搜索
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG+AND+ti:attention&max_results=10&sortBy=submittedDate',
prompt: '查找相关论文,提取标题和摘要',
});
// 该领域的最新论文
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG&max_results=20&sortBy=submittedDate&sortOrder=descending',
prompt: '提取 20 篇最新的机器学习论文',
});
</execution_process>
<best_practices>
sortBy=submittedDate&sortOrder=descending</best_practices>
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=ti:transformer+AND+abs:attention&max_results=10&sortBy=relevance',
prompt: '提取论文标题、作者、摘要和 arXiv ID',
});
示例 2:查找研究人员的论文:
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=au:Vaswani&max_results=15',
prompt: '列出该作者的所有论文,包含标题和日期',
});
示例 3:获取最新的机器学习论文:
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG&max_results=20&sortBy=submittedDate&sortOrder=descending',
prompt: '提取 20 篇最新的机器学习论文,包含标题和摘要',
});
示例 4:使用 Exa 进行语义搜索:
mcp__Exa__web_search_exa({
query: 'site:arxiv.org multimodal large language models vision 2024',
numResults: 10,
});
示例 5:获取特定论文详情:
WebFetch({
url: 'http://export.arxiv.org/api/query?id_list=1706.03762',
prompt: "提取 'Attention Is All You Need' 论文的完整详情",
});
</usage_example>
此技能自动分配给:
search_query=neural+networks 会返回数千条结果;始终使用 ti:、au:、cat: 或 abs: 前缀来限定查询范围。| 反模式 | 失败原因 | 正确方法 |
|---|---|---|
使用 max_results=100 或无限制 | 上下文爆炸;100 篇论文 × 300 字节 = 30KB+ 元数据 | 始终设置 max_results=20(硬性限制) |
| 获取完整论文 PDF | 单篇论文可能超过 100KB;耗尽上下文预算 | 仅通过 API 提取摘要 + 元数据 |
| 不带字段前缀的宽泛查询 | 返回所有字段的不相关结果 | 使用 ti:、au:、cat: 或 abs: 前缀 |
| 仅使用 WebFetch 进行发现 | 错过语义相关但不完全匹配确切术语的论文 | 首先使用 Exa 进行语义发现 |
| 引用论文标题而非 arXiv ID | 标题可能具有歧义或重复 | 始终包含 arXiv ID(例如 1706.03762) |
开始前:
cat .claude/context/memory/learnings.md
完成后:
.claude/context/memory/learnings.md.claude/context/memory/issues.md.claude/context/memory/decisions.md假设可能中断:您的上下文可能会重置。如果它不在内存中,则表示未发生。
每周安装次数
82
仓库
GitHub 星标数
19
首次出现
2026 年 1 月 29 日
安全审计
安装于
github-copilot81
gemini-cli80
cursor80
kimi-cli79
amp79
codex79
Mode: Cognitive/Prompt-Driven — No standalone utility script; use via agent context.
This skill uses existing tools to access arXiv:
Works immediately - no MCP server, no restart needed.
arxiv-mcp returns academic papers. To prevent memory exhaustion:
Why the limit?
The arXiv API is publicly accessible at http://export.arxiv.org/api/query.
// ✓ GOOD: Limit results to 20
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:transformer+attention&max_results=20&sortBy=relevance',
prompt: 'Extract paper titles, authors, abstracts, arXiv IDs, and PDF links from these results',
});
// ✓ GOOD: Use specific filters to reduce result set
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:transformer+attention+2025&max_results=20&sortBy=submittedDate',
prompt: 'Extract recent papers on transformer attention',
});
// ✗ BAD: Old behavior - unlimited or >20 results
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:neural+networks',
// Too broad - will get 100s of results
});
// ✗ BAD: Exceeds memory limit
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:deep+learning&max_results=100',
// Over limit - memory risk
});
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=all:transformer+attention&max_results=20&sortBy=relevance',
prompt: 'Extract paper titles, authors, abstracts, arXiv IDs, and PDF links from these results',
});
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=au:LeCun&max_results=10&sortBy=submittedDate',
prompt: 'Extract paper titles, authors, abstracts, and arXiv IDs',
});
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG&max_results=15&sortBy=submittedDate',
prompt: 'Extract paper titles, authors, abstracts, categories, and arXiv IDs',
});
WebFetch({
url: 'http://export.arxiv.org/api/query?id_list=2301.07041',
prompt:
'Extract full details: title, all authors, abstract, categories, published date, PDF link',
});
| Parameter | Description | Example |
|---|---|---|
search_query | Search terms with field prefixes | all:transformer, au:LeCun, ti:attention |
id_list | Comma-separated arXiv IDs | 2301.07041,2302.13971 |
max_results | Number of results (default 10, max 100) |
| Prefix | Field | Example |
|---|---|---|
all: | All fields | all:machine+learning |
ti: | Title | ti:transformer |
au: | Author | au:Vaswani |
abs: |
Combine terms with AND, OR, ANDNOT:
search_query=ti:transformer+AND+abs:attention
search_query=au:LeCun+OR+au:Bengio
search_query=cat:cs.LG+ANDNOT+ti:survey
pnpm search:code or ripgrep skill on codebase (Grep/Glob as fallback)arxiv-mcp is best for:
Use Exa for more natural language queries with arXiv filtering:
mcp__Exa__web_search_exa({
query: 'site:arxiv.org transformer architecture attention mechanism deep learning',
numResults: 10,
});
mcp__Exa__web_search_exa({
query: 'site:arxiv.org large language model scaling laws 2024',
numResults: 15,
});
mcp__Exa__web_search_exa({
query: 'site:arxiv.org author:"Yann LeCun" deep learning',
numResults: 10,
});
| Category | Field |
|---|---|
| cs.AI | Artificial Intelligence |
| cs.LG | Machine Learning |
| cs.CL | Computation and Language (NLP) |
| cs.CV | Computer Vision |
| cs.SE | Software Engineering |
| cs.CR | Cryptography and Security |
| stat.ML | Machine Learning (Statistics) |
| math.* | Mathematics (all subcategories) |
| physics.* | Physics (all subcategories) |
| q-bio.* | Quantitative Biology |
| econ.* | Economics |
// Start with broad Exa search for semantic matching
mcp__Exa__web_search_exa({
query: 'site:arxiv.org transformer attention mechanism neural networks',
numResults: 10,
});
// Get details for interesting papers by ID
WebFetch({
url: 'http://export.arxiv.org/api/query?id_list=2301.07041,2302.13971',
prompt: 'Extract full metadata for each paper: title, authors, abstract, categories, PDF URL',
});
// Search by category of interesting paper
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG+AND+ti:attention&max_results=10&sortBy=submittedDate',
prompt: 'Find related papers, extract titles and abstracts',
});
// Latest papers in the field
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG&max_results=20&sortBy=submittedDate&sortOrder=descending',
prompt: 'Extract the 20 most recent machine learning papers',
});
</execution_process>
<best_practices>
sortBy=submittedDate&sortOrder=descending</best_practices>
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=ti:transformer+AND+abs:attention&max_results=10&sortBy=relevance',
prompt: 'Extract paper titles, authors, abstracts, and arXiv IDs',
});
Example 2: Find papers by researcher :
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=au:Vaswani&max_results=15',
prompt: 'List all papers by this author with titles and dates',
});
Example 3: Get recent ML papers :
WebFetch({
url: 'http://export.arxiv.org/api/query?search_query=cat:cs.LG&max_results=20&sortBy=submittedDate&sortOrder=descending',
prompt: 'Extract the 20 most recent machine learning papers with titles and abstracts',
});
Example 4: Semantic search with Exa :
mcp__Exa__web_search_exa({
query: 'site:arxiv.org multimodal large language models vision 2024',
numResults: 10,
});
Example 5: Get specific paper details :
WebFetch({
url: 'http://export.arxiv.org/api/query?id_list=1706.03762',
prompt: "Extract complete details for the 'Attention Is All You Need' paper",
});
</usage_example>
This skill is automatically assigned to:
search_query=neural+networks returns thousands of results; always scope with ti:, au:, cat:, or abs: prefixes to target the query.| Anti-Pattern | Why It Fails | Correct Approach |
|---|---|---|
Using max_results=100 or no limit | Context explosion; 100 papers × 300 bytes = 30KB+ metadata | Always set max_results=20 (hard limit) |
| Fetching full paper PDFs | Single paper can be 100KB+; kills context budget | Extract abstract + metadata only via API |
| Broad query without field prefix | Returns irrelevant results across all fields | Use ti:, au:, cat:, or abs: prefix |
Before starting:
cat .claude/context/memory/learnings.md
After completing:
.claude/context/memory/learnings.md.claude/context/memory/issues.md.claude/context/memory/decisions.mdASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.
Weekly Installs
82
Repository
GitHub Stars
19
First Seen
Jan 29, 2026
Security Audits
Gen Agent Trust HubPassSocketPassSnykFail
Installed on
github-copilot81
gemini-cli80
cursor80
kimi-cli79
amp79
codex79
AI 代码实施计划编写技能 | 自动化开发任务分解与 TDD 流程规划工具
49,800 周安装
max_results=20 |
start | Offset for pagination | start=10 |
sortBy | Sort order: relevance, lastUpdatedDate, submittedDate | sortBy=submittedDate |
sortOrder | ascending or descending | sortOrder=descending |
| Abstract |
abs:attention+mechanism |
cat: | Category | cat:cs.LG |
co: | Comment | co:accepted |
| Using only WebFetch for discovery | Misses semantically related papers not matching exact terms | Use Exa for semantic discovery first |
| Citing paper titles instead of arXiv IDs | Titles can be ambiguous or duplicated | Always include the arXiv ID (e.g., 1706.03762) |