arXiv Search

Search arXiv preprints by keyword and category: title, abstract, authors, dates, categories, DOI and PDF link, by relevance or date.

Pay by wallet (x402) arxiv-search

Example

Example answer

{
  "query": "transformer attention",
  "papers": [
    {
      "doi": null,
      "title": "Dilated Neighborhood Attention Transformer",
      "abs_url": "http://arxiv.org/abs/2209.15001v3",
      "authors": [
        "Ali Hassani"
      ],
      "pdf_url": "https://arxiv.org/pdf/2209.15001v3",
      "summary": "Transformers are quickly becoming one of the most heavily applied deep learning architectures...",
      "updated": "2023-01-16T18:58:58Z",
      "version": 3,
      "arxiv_id": "2209.15001",
      "published": "2022-09-29T17:57:08Z",
      "categories": [
        "cs.CV"
      ],
      "primary_category": "cs.CV"
    }
  ],
  "category": null,
  "returned": 1,
  "total_results": 258539
}

Inputs

Field	Type	Required	What it is
`sort`	string	No	relevance (default), date (newest submitted first) or updated (most recently revised first)
`limit`	integer	No	Results to return, 1-25. Default: 10
`query`	string	Yes	Free-text search across title, abstract and authors. At least 2 characters.
`category`	string	No	Restrict to an arXiv category such as cs.CL, stat.ML or astro-ph.GA

Output fields

How reliably each field appears in a successful answer.

Field	Type	Appears
`query`	string	Always (present in every successful answer)
`papers`	array	Always (present in every successful answer)
`returned`	number	Always (present in every successful answer)
`total_results`	number	Always (present in every successful answer)
`category`	—	Usually (present for most inputs)

How to call it

The same tool through each route. MCP clients reach it through Strale’s server at https://api.strale.io/mcp; the call below is what the agent sends.

# Needs an API key. max_price_cents caps what this call may cost.
curl -X POST https://api.strale.io/v1/do \
  -H "Authorization: Bearer $STRALE_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
    "capability_slug": "arxiv-search",
    "inputs": {
      "query": "attention is all you need",
      "limit": 3
    },
    "max_price_cents": 2
  }'

{
  "tool": "strale_execute",
  "arguments": {
    "slug": "arxiv-search",
    "inputs": {
      "query": "attention is all you need",
      "limit": 3
    },
    "max_price_cents": 2
  }
}

import { Strale } from "straleio";

const strale = new Strale({ apiKey: process.env.STRALE_API_KEY! });
const result = await strale.do({
  capability_slug: "arxiv-search",
  inputs: {
    "query": "attention is all you need",
    "limit": 3
  },
  max_price_cents: 2,
});

import os
from straleio import Strale, DoRequest

strale = Strale(api_key=os.environ["STRALE_API_KEY"])
result = strale.do(DoRequest(
    capability_slug="arxiv-search",
    inputs={
        "query": "attention is all you need",
        "limit": 3,
    },
    max_price_cents=2,
))

# No account. The first request answers 402 with the price in USDC;
# an x402 client pays it from your wallet and repeats the request.
curl -i -X POST https://api.strale.io/x402/v2/arxiv-search \
  -H 'Content-Type: application/json' \
  -d '{
    "query": "attention is all you need",
    "limit": 3
  }'

Limitations

arXiv asks clients to space requests about 3 seconds apart
Bursts above roughly 20 requests per minute are answered with HTTP 429/503 and surfaced as a retryable upstream error; the platform does not queue on the caller's behalf.
What to do instead: Space calls out or batch several terms into one query.
Short queries can return loosely ranked results
One- or two-word queries return large, loosely ranked result sets; total_results is arXiv's estimate.
What to do instead: Add a category filter or use sort=date for recency.