Semantic Scholar MCP Server: Papers and Citations

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → Semantic Scholar, install it with no key, mint an MCP API key and you're done. No Docker, no git clone, no local server.

Semantic Scholar

Scholarly literature from Semantic Scholar, no key needed: search more than 200 million papers with abstracts and AI summaries, resolve titles and DOIs, follow citations and references with their citing sentences, read author profiles with h-index, get paper recommendations and search full-text snippets. Read-only; a free key adds its own quota.

Setup

Works without a key: setup takes under a minute. The connector reads the Semantic Scholar Academic Graph API of the Allen Institute for AI (https://api.semanticscholar.org). It answers without an account or key, from a pool that all keyless callers share, so busy hours often bring 429 (Too Many Requests).

1. In the connector store, pick Semantic Scholar and install it. There is no Authorize with Provider step.
2. Recommended: request a free key with the form at https://www.semanticscholar.org/product/api#api-key-form. Semantic Scholar reviews each request, usually within days. Paste the key into SEMANTIC_SCHOLAR_API_KEY: it is sent as the x-api-key header and gives the connector its own quota, starting at 1 request per second.
3. Check it with semantic_scholar_list_dataset_releases: it lists the release dates of the bulk datasets.

Paper and author ids: every paper tool accepts the 40-character Semantic Scholar id or a prefixed external id: DOI:10.18653/v1/N18-3011, ARXIV:2106.15928, PMID:19872477, PMCID:2323736, CorpusId:215416146, MAG:112218234, ACL:W12-3903 or URL:https://arxiv.org/abs/2106.15928. Author ids are numbers such as 1741101.

What it reads:

• More than 200 million scholarly papers from all fields: relevance search with plain words, filtered by year or date, venue, field of study, publication type, open access PDF and minimum citations; bulk search with boolean syntax, sorted by citations or date; the best match for a title, with its matchScore; suggestions for a partial title.
• One paper: title, abstract, AI summary (tldr), authors, venue, year, citation and reference counts, open access PDF and external ids. Up to 500 papers in one call.
• The citation graph: who cites a paper (citingPaper) and what it cites (citedPaper), with the citing sentences (contexts), the reason (intents: background, methodology, result) and isInfluential for citations that build on the work.
• Authors: search by name, profile with affiliations, homepage, paper and citation counts and h-index, their papers in a date range, up to 1,000 authors in one call. Several profiles can share a name, so the AI compares affiliations and paper counts before choosing.
• Recommendations from one seed paper, or from a list of liked and optionally disliked papers.
• Full-text snippets: passages of about 500 words with section and score, from open access papers whose text Semantic Scholar holds.
• The releases of the bulk datasets and the datasets each one contains.

Data formats: the fields parameter chooses what comes back, comma-separated without spaces; each tool has a compact default. Nested fields use a dot, e.g. authors.affiliations. abstract, tldr and citationStyles (BibTeX) make answers large, so the AI asks for them only when needed. year is the publication year and publicationDate is YYYY-MM-DD when known. citationCount counts all citing papers, influentialCitationCount only the influential ones. openAccessPdf.url is a free PDF when present, and externalIds holds DOI, ArXiv, PubMed and other ids. Cite Semantic Scholar and the paper's DOI or URL.

Paging: list answers carry offset and next; the AI calls again with offset set to next until next is missing. Relevance search returns up to 100 papers per page and reaches at most 1,000 results. Bulk search returns up to 1,000 papers per call and pages with token through up to 10 million.

Read-only: every tool only reads (the batch and recommendation tools are reads sent as POST), so there is nothing to confirm and no tool is switched off at install.

Limits: without a key the shared pool answers 429 when it is busy: the AI waits a few seconds and tries again, or you add a key. With a key the default quota is 1 request per second.

Common errors: 400 names a wrong field or parameter. 404 means an unknown paper or author id, or "Title match not found" when no title is close. 403 means the key is wrong or disabled.

The connector does not download the bulk dataset files (they need a key and are many gigabytes) and does not return the per-paper embedding vectors.

Prerequisites: (none required; optional: SEMANTIC_SCHOLAR_API_KEY)

Install the connector locally

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Open http://localhost:3000/connectors/store, pick Semantic Scholar and install it. The optional key can stay empty.

Available tools

Example prompts

• "Find the most cited papers on retrieval-augmented generation since 2022 and summarise each in one sentence."
• "Which papers cite "Attention Is All You Need" for its methodology, and what do the citing sentences say?"
• "Show the h-index, affiliation and five most cited papers of the author Yoshua Bengio."
• "I liked these three papers on protein structure prediction: recommend ten more I should read."
• "Find passages in open access papers that discuss learning rate warmup and training stability."

Next steps

• Connect to Claude
• Connect to ChatGPT