Ask for paperId, title, year and citationCount and a twenty-result search stays small; ask for the default record and every abstract comes with it. get_citation then formats the reference in BibTeX, APA, MLA or Chicago, which is the step people otherwise do by hand.
A Python server over the Semantic Scholar API. Four tools, each taking a fields parameter so you request only the columns you need rather than the whole record.
- Paper search filtered by year, fields of study and open-access PDF availability — search_paper
- One paper by any of its identifiers — Semantic Scholar id, DOI, arXiv id, MAG, ACL, PubMed or Corpus id — get_paper
- A paper's authors with affiliations, citation counts and h-index — get_authors
- A citation formatted as BibTeX, APA, MLA or Chicago — get_citation
- Field selection on every tool, so a search can return just paperId, title, year and citationCount instead of abstracts and reference lists
Runs with uv straight from the repository: uv run --with git+https://github.com/FujishigeTemma/semantic-scholar-mcp semantic-scholar-mcp serve. SEMANTIC_SCHOLAR_API_KEY is optional and raises the rate limit — 100 requests per 5 minutes without one, 1 request per second with.
One command — claude mcp add semantic-scholar-mcp -s project -e SEMANTIC_SCHOLAR_API_KEY="your-api-key-here" -- uv run --with "git+https://github.com/FujishigeTemma/semantic-scholar-mcp" semantic-scholar-mcp serve