| name | arxiv-search |
| description | Search arXiv for latest AI/ML preprints. Use when Claude needs bleeding-edge research not yet in peer-reviewed journals — newest methods, architectures, training techniques, benchmarks. |
| keywords | ["arxiv","preprints","papers","research","machine-learning","deep-learning","AI","foundation-models"] |
arXiv Search
Search arXiv preprints across AI/ML categories: cs.AI, cs.LG, cs.CL, cs.CV, stat.ML, cs.MA.
No API key required. Uses arXiv public API directly.
Usage
Find the script path, then execute:
SCRIPT=$(find ~/.claude/plugins -path "*/ai-frontier/skills/arxiv-search/scripts/search.mjs" 2>/dev/null | head -1)
node "$SCRIPT" "your query" [maxResults] [--sort=date|relevance] [--cats=cs.AI,cs.LG]
Arguments
| Arg | Default | Description |
|---|
| query | (required) | Natural language search query |
| maxResults | 10 | Number of results (1-100) |
| --sort=date | relevance | Sort by submission date or relevance |
| --cats=cs.AI,cs.LG | all 6 defaults | Comma-separated arXiv categories |
When to Use
- Finding the newest preprints on a topic (use
--sort=date)
- Searching for specific methods or architectures
- Getting abstracts and PDF links for deeper reading
- Checking if new work exists in a specific arXiv category
Output Format
Returns JSON with unified envelope:
results[].title — paper title
results[].authors — author list
results[].abstract — first 500 chars of abstract
results[].pdf_url — direct PDF link
results[].source_specific.arxiv_id — arXiv identifier
results[].source_specific.categories — arXiv categories
results[].source_specific.published — publication date
Examples
node "$SCRIPT" "chain-of-thought reasoning" 10 --sort=date
node "$SCRIPT" "diffusion models" 5 --cats=cs.CV
node "$SCRIPT" "AI alignment safety" 20