Academic Search — Paper Discovery
Overview
Search academic papers on arXiv using the paper-search CLI tool. No API key required — uses the free arXiv API directly.
When To Use
-
User asks to find/search papers on a topic
-
User wants paper recommendations
-
User needs to download a paper PDF
-
User mentions arXiv, academic papers, or research literature
Commands
Search papers by keyword
paper-search search "transformer attention mechanism" --max 5
Search with category filter
paper-search search "reinforcement learning" --max 10 --categories cs.AI,cs.LG
Get paper details by arXiv ID
paper-search details 1706.03762
Download paper PDF
paper-search download 1706.03762 --output /workspace/papers/
JSON output (for programmatic use)
paper-search search "neural networks" --max 3 --json
Example Workflows
Find important papers on a topic
Search for recent transformer papers
paper-search search "vision transformer ViT" --max 10 --categories cs.CV
Get details of the seminal paper
paper-search details 2010.11929
Build a reading list
Search and save results
paper-search search "large language models" --max 20 --json > /workspace/papers/llm-papers.json
Download key papers
paper-search download 2005.14165 -o /workspace/papers/ # GPT-3 paper-search download 2302.13971 -o /workspace/papers/ # LLaMA
Research a specific author
paper-search search "au:Hinton" --max 10
arXiv Categories
Common categories for filtering:
Category Description
cs.AI Artificial Intelligence
cs.LG Machine Learning
cs.CL Computation and Language (NLP)
cs.CV Computer Vision
cs.NE Neural and Evolutionary Computing
cs.RO Robotics
stat.ML Statistics - Machine Learning
math.OC Optimization and Control
Search Tips
-
Use quotes for exact phrases: "attention is all you need"
-
Search by author: au:Vaswani
-
Search by title: ti:transformer
-
Combine: "neural network" AND au:LeCun
-
arXiv ID format: 1706.03762 or cs.AI/0001001
Output Locations
-
Papers list: stdout or --json to file
-
Downloaded PDFs: /workspace/papers/ (default)
Important Notes
-
arXiv API has rate limits; avoid rapid-fire requests
-
PDFs are typically 1-10 MB each
-
Search results are sorted by relevance by default
-
Published dates are in YYYY-MM-DD format