semantic-scholar-api

name	semantic-scholar-api
description	Use the Semantic Scholar Graph API to discover relevant computer vision papers, citation counts, abstracts, and paper metadata before implementing or tuning algorithms.
allowed-tools	read_file run_python_inline
compatibility	Requires outbound HTTPS access. Optional SEMANTIC_SCHOLAR_API_KEY improves rate limits.

Semantic Scholar API

Use this skill when the task needs paper discovery, metadata lookup, citation filtering, or related-work triage before implementing a CV solution.

Workflow

1. Use run_python_inline to query the Semantic Scholar Graph API.
2. Start with paper search for task phrases from the prompt markdown.
3. Narrow by year, venue, citation count, and fields of study when the result set is large.
4. Save the most relevant titles, paper IDs, abstracts, and URLs into workspace notes before acting on them.
5. Use arXiv separately when you need the full text.

Python Pattern

import json
import os
import requests

API_KEY = os.getenv("SEMANTIC_SCHOLAR_API_KEY")
headers = {"Accept": "application/json"}
if API_KEY:
    headers["x-api-key"] = API_KEY

params = {
    "query": "translation-only image registration benchmark",
    "limit": 10,
    "fields": "title,abstract,year,venue,url,citationCount,authors",
}
response = requests.get(
    "https://api.semanticscholar.org/graph/v1/paper/search",
    params=params,
    headers=headers,
    timeout=30,
)
response.raise_for_status()
payload = response.json()
print(json.dumps(payload, indent=2))

Useful Endpoints

• Search papers: GET https://api.semanticscholar.org/graph/v1/paper/search
• Fetch a paper by paper ID, DOI, arXiv ID, or URL-derived identifier: GET https://api.semanticscholar.org/graph/v1/paper/{paper_id}

Recommended Fields

For search:

• title
• abstract
• year
• venue
• url
• citationCount
• authors
• externalIds

For follow-up fetches:

• referenceCount
• influentialCitationCount
• fieldsOfStudy
• openAccessPdf

Practical Notes

• Favor papers with abstracts and reproducible baselines.
• Use multiple queries: task name, modality, and artifact type.
• If the task is very applied, combine this skill with Tavily for codebases or benchmarks.