| name | semantic-scholar |
| description | How to use the semantic-scholar MCP tools for academic paper lookup, citation tracing, author profiling, and paper recommendations. Use this skill whenever the user asks about academic papers, citations, references, author h-index, paper metadata, or wants to search the academic literature — even if they don't mention "Semantic Scholar" by name. Also use it when you need to look up a paper by arXiv ID, DOI, or title. |
Semantic Scholar MCP Tools
This skill covers the 8 tools provided by the semantic-scholar MCP server. These tools give you access to 200M+ academic papers via the Semantic Scholar Academic Graph API.
Important: these tools return metadata only — title, abstract, authors, year, citation counts, and paper IDs. They do not return full paper text. If the user needs to read a paper's methodology, results, or analysis in detail, you'll need a separate full-text tool (like alphaxiv).
Tools at a Glance
| Tool | What it does | When to use |
|---|
paper | Single paper metadata | You have a paper ID and need its details |
paperBatch | Up to 500 papers at once | You have multiple IDs to look up |
relevanceSearch | Keyword search with filters | Finding papers on a topic |
citations | Papers that cite a given paper | Tracing forward influence |
references | Papers cited by a given paper | Tracing backward lineage |
recommendations | Similar papers from seed papers | Discovering related work |
author | Author profile | Getting h-index, paper count, affiliations |
authorPapers | Author's publication list | Browsing someone's body of work |
Paper ID Formats
All tools that accept a paper_id understand these formats:
| Format | Example | Notes |
|---|
| ARXIV (prefixed) | ARXIV:1706.03762 | Most reliable for arXiv papers |
| ARXIV (bare) | 2005.14165 | Auto-prefixed — works the same |
| DOI (prefixed) | DOI:10.1038/s41586-021-03819-2 | For journal papers |
| DOI (bare) | 10.1038/s41586-021-03819-2 | Auto-prefixed |
| S2 ID | 204e3073870fae3d05bcbc2f6a8e263d9b72e776 | Semantic Scholar's internal hex ID |
| PMID | PMID:34265844 | PubMed papers |
| URL | URL:https://arxiv.org/abs/1706.03762 | Any paper URL |
When the user gives you an arXiv link like https://arxiv.org/abs/2301.12345, extract the ID and use ARXIV:2301.12345.
Tool Details
relevanceSearch
The go-to tool for finding papers on a topic. Returns ranked results with metadata.
Parameters:
query (required) — keywords or paper title
limit — max results (default 10, max 100)
offset — for pagination
year — range filter: "2023-2024", "2020-", "-2023"
fields_of_study — e.g. "Computer Science", "Mathematics", "Physics"
min_citation_count — integer threshold (useful for finding impactful work)
open_access_only — boolean, filters to papers with free PDFs
Tips:
- Use specific technical terms rather than broad queries
- Combine
year + min_citation_count to find recent high-impact work
- The
total field in the response tells you how many papers match overall
paper / paperBatch
Look up known papers by ID. paperBatch is more efficient when you have multiple IDs — it returns null for any ID not found, without failing the whole request.
Returns: title, abstract, year, authors (with IDs), citationCount, influentialCitationCount, referenceCount, externalIds, fieldsOfStudy, venue, publicationDate, openAccessPdf, tldr.
citations / references
Traverse the citation graph. citations gives you papers that cite the target (forward in time). references gives you papers the target cites (backward in time).
Parameters:
paper_id (required)
limit — default 100, max 1000
offset — for pagination through large citation lists
Returns include:
contexts — the actual sentences where the citation appears
intents — why it was cited (background, methodology, result)
isInfluential — whether this is a substantive citation vs. a passing mention
Use offset + limit to page through highly-cited papers (e.g., "Attention Is All You Need" has 175k+ citations).
recommendations
Finds papers similar to your seed papers. Supports steering with negative seeds.
Parameters:
positive_paper_ids (required) — papers you want more of
negative_paper_ids (optional) — papers to steer away from
limit — default 100, max 500
This is powerful for literature discovery when you have a few anchor papers and want to expand. Negative seeds help when recommendations drift into an adjacent but unwanted subfield.
author / authorPapers
Look up researcher profiles and their publications.
author takes a numeric author ID (found in paper results under authors[].authorId)
authorPapers returns their publication list with pagination
Common Workflows
Find high-impact recent papers on a topic:
relevanceSearch(query, year="2023-", min_citation_count=50, limit=20)
Get a paper and its most important references:
paper(id) → then references(id, limit=20) → filter by isInfluential=true
Trace how a paper influenced the field:
citations(id, limit=50) → sort by citationCount → look at top citers
Discover related work from anchor papers:
recommendations(positive=[paper_a, paper_b], negative=[unrelated_paper], limit=20)
Profile a researcher:
author(id) → authorPapers(id, limit=50) → sort by citationCount
Things to Keep in Mind
- Rate limits: Without an API key, you get 1 request/second. With a key, 100 req/s. If you're doing batch operations, prefer
paperBatch over multiple paper calls.
- Metadata only: These tools don't give you full paper text. Don't summarize a paper based solely on its abstract — tell the user you only have the abstract if that's all you've read.
- Author IDs are numeric: You'll find them in paper results. They're not names — you need the ID to call
author or authorPapers.
- Pagination: For
citations/references/authorPapers, use offset and limit. The response includes a next field indicating the offset for the next page.
- Null in batch:
paperBatch returns null for papers not found — this is normal, not an error.