arXiv Search Skill

Search and retrieve metadata for preprints on arXiv, the open-access repository for scholarly articles.

Critical: Things Claude Won't Know Without This Skill

Response format is Atom XML, not JSON

The arXiv API returns Atom 1.0 XML — there is no JSON option. You must parse XML to extract results. The Python script handles this automatically.

The base URL uses export.arxiv.org, not arxiv.org

http://export.arxiv.org/api/query?search_query=all:transformer&max_results=5

Using arxiv.org directly will not work for API queries.

Boolean operators must be UPPERCASE

Use AND, OR, ANDNOT — lowercase will not work:

search_query=ti:attention AND ti:transformer
search_query=cat:cs.AI ANDNOT cat:cs.CL

Parentheses and quotes must be URL-encoded

• Parentheses: %28 and %29
• Quotes (phrase search): %22

search_query=ti:%22large+language+model%22
search_query=au:bengio+AND+%28cat:cs.LG+OR+cat:cs.AI%29

Date range filtering uses a special syntax

search_query=submittedDate:[202401010000+TO+202412312359]

Format: YYYYMMDDTHHMM in GMT, 24-hour. The T is literal. Combine with other queries using AND.

Rate limiting: 3 seconds between requests

arXiv asks for a minimum 3-second delay between API calls. Results update once daily, so there is no reason to poll frequently.

Hard limits on results

• max_results caps at 2000 per request
• Total retrievable results cap at 30,000 (start + max_results ≤ 30000)
• Exceeding 30,000 returns HTTP 400

Search Field Prefixes

Python Script

Use scripts/arxiv_api.py for programmatic access (zero dependencies):

from scripts.arxiv_api import ArxivAPI
api = ArxivAPI()

# Simple keyword search
results = api.search("all:transformer attention mechanism", max_results=10)

# Field-specific search
results = api.search("ti:diffusion AND cat:cs.CV", max_results=5)

# Author search
results = api.search("au:hinton AND cat:cs.LG", max_results=20)

# Date-filtered search
results = api.search(
    "cat:cs.AI AND submittedDate:[202401010000 TO 202412312359]",
    sort_by="submittedDate",
    sort_order="descending",
    max_results=50
)

# Fetch specific papers by arXiv ID
results = api.fetch_by_ids(["2301.07041", "2303.08774", "2305.10601"])

# Paginated retrieval
page1 = api.search("cat:cs.CL", max_results=100, start=0)
page2 = api.search("cat:cs.CL", max_results=100, start=100)

# Each result is a dict with keys:
# id, title, summary, authors, published, updated,
# categories, primary_category, links, comment, journal_ref, doi

Common Category Codes

Full list: https://arxiv.org/category_taxonomy

API Etiquette

• Rate limit: 3-second minimum between requests
• No authentication required
• Results update daily — no need to poll the same query more than once per day
• Use pagination for large result sets instead of requesting thousands at once
• Respect HTTP 429: Back off if rate-limited

Related Skills

• wikidata-search: Cross-reference arXiv authors with Wikidata entities for authority control identifiers (ORCID, VIAF, etc.)
• harvard-library-catalog: Look up published versions of arXiv preprints in Harvard's bibliographic records

Resources

• references/api_reference.md — Complete API specs for query parameters, response fields, and error handling
• scripts/arxiv_api.py — Python client with XML parsing, rate limiting, and pagination support