Search Semantic Scholar for papers matching a query.

Usually completes in a few seconds.

Answers directly in normal use. Should the call run long it continues in the background and returns a job handle to poll with get_job_result.

Fetch full metadata for a single paper.

A cached paper answers immediately. Otherwise the record is fetched and enriched; should that run long it continues in the background and returns a job handle to poll with get_job_result.

Fetch author profile and publications, or search by name.

If identifier looks like a numeric S2 author ID, fetches the author directly. Otherwise performs a name search and returns up to 5 candidates for disambiguation.

Answers directly in normal use. Should the call run long it continues in the background and returns a job handle to poll with get_job_result.

Fetch papers that cite the given paper (forward citations).

Paging deeply to satisfy min_citations can run long; such a call continues in the background and returns a job handle to poll with get_job_result.

Fetch papers referenced by the given paper (backward references).

Should the call run long it continues in the background and returns a job handle to poll with get_job_result.

Traverse the citation graph from one or more seed papers.

Performs BFS up to depth hops. Returns nodes (paper records) and directed edges. Hard-caps at max_nodes to prevent runaway expansion.

Traversal makes one rate-limited request per node, so anything past a shallow, narrow graph runs long: expect a job handle to poll with get_job_result rather than the graph itself.

A failed upstream request does not abort the walk. When one happens, stats.partial is true, stats.failed_requests counts them, and a warning key explains what went unfetched. Treat such a graph as incomplete: a missing node or edge is unknown, not absent. This is distinct from stats.truncated, which means max_nodes stopped an otherwise successful walk.

Find the shortest citation path between two papers.

Uses BFS over the citation/reference graph. Leverages cached citation and reference lists to minimise API calls.

The search walks outward one rate-limited request per node, so it commonly runs long and returns a job handle to poll with get_job_result rather than the path itself.

A failed upstream request does not abort the search. When one happens, partial is true, failed_requests counts them, and a warning key explains what went unfetched. A partial {"found": false} is not evidence that no path exists, and a partial {"found": true} path is not guaranteed to be the shortest one.

Recommend papers based on positive (and optionally negative) examples.

Answers directly in normal use. Should the call run long it continues in the background and returns a job handle to poll with get_job_result.

Resolve a list of paper, patent, or book identifiers to full records.

Uses the S2 batch endpoint for paper IDs/DOIs, with OpenAlex fallback. Patent numbers (e.g. EP1234567A1) are auto-detected and resolved via the EPO OPS API when configured. ISBNs (prefixed ISBN:) are resolved via Open Library.

Identifiers fan out across three upstreams, so a large batch runs long and is then handed back as a job handle to poll with get_job_result rather than the records themselves.

Fetch OpenAlex metadata to supplement Semantic Scholar data.

Resolves the paper's DOI from S2, then queries OpenAlex for the requested enrichment fields. Results are cached for 30 days.

A cached record answers immediately. Should the lookup run long it continues in the background and returns a job handle to poll with get_job_result.

Generate formatted citations for one or more papers.

Resolves papers via Semantic Scholar, optionally enriches with OpenAlex metadata, and formats as BibTeX, CSL-JSON, or RIS.

Enrichment fans out per paper, so a large batch can take a while; such a call answers with a job handle to poll using get_job_result instead of the citations themselves.

Search for books by title, author, or free text.

Uses Open Library. Prefer title and author over query — they use dedicated indexes and return far better results.

Open Library is politeness-rate-limited, so a busy moment may answer with a job handle to poll using get_job_result instead of the results themselves.

Fetch book metadata by ISBN or Open Library ID.

Open Library is politeness-rate-limited, so a busy moment may answer with a job handle to poll using get_job_result instead of the record itself.

Get a book excerpt and preview info from Google Books.

Returns the publisher description, text snippet, and a link to the Google Books preview page. Google Books does not expose full chapter text via API -- the excerpt is a publisher-provided summary and/or search snippet.

An error of not_found means Google Books answered and has no such volume: stop asking about that ISBN. A lookup that never got an answer says so instead, as rate_limited (with retryable) or upstream_error, and is worth retrying -- the book may well exist.

Recommend books for a subject via Open Library.

Uses the Open Library subject API to find popular books on a topic, sorted by edition count (a proxy for popularity).

Normalise a messy standard citation string to its canonical form.

Tries local regex first (fast, no network). Falls back to querying source APIs when local patterns don't match. Returns all candidates when the input is ambiguous.

A cold catalogue makes this slow: the first call after a fresh install or a cleared cache downloads and parses each body's index, which runs well past the soft deadline. Such a call returns a job handle to poll with get_job_result rather than the result itself.

Search technical standards by identifier, title, or free text.

Searches NIST, IETF, W3C, and ETSI. Use body to restrict to one source body.

A cold catalogue makes this slow: the first call after a fresh install or a cleared cache downloads and parses each body's index, which runs well past the soft deadline. Such a call returns a job handle to poll with get_job_result rather than the result itself.

Retrieve a standard by identifier (canonical or fuzzy).

Resolves fuzzy inputs (e.g. "rfc9000", "nist 800-53") to their canonical form before fetching. With fetch_full_text=True and docling configured, downloads and converts the full text.

With fetch_full_text=True the docling conversion takes minutes, so such a call returns a job handle to poll with get_job_result rather than the record itself. Without it, and on a cache hit, the record comes back directly.

A conversion that fails still returns the record, with the reason in full_text_error; full_text_url is there to fetch by hand. No full_text and no full_text_error means none was on offer, or docling is not configured — neither is worth retrying.

Report the last sync run for each standards body.

One row per body. started_at / finished_at are Unix timestamps (seconds). errors is a list of non-fatal error strings from the most recent run (empty on success).

Returns: {"runs": [{body, upstream_ref, added, updated, unchanged, withdrawn, errors, started_at, finished_at}, ...]}. Empty runs list when no sync has been run yet.

Retrieve the outcome of a background job started by a long-running tool on this server. When such a tool answers with status "working" and a job_id, call this tool with that job_id every few seconds until the status is terminal. Job records expire after a while — fetch results soon after completion. Scholar MCP promotes slow work to a background job. PDF download and docling conversion usually take 1-5 minutes, a busy EPO traffic light is waited out, citation formatting enriches each paper in turn, and graph traversal makes one rate-limited request per node. Those calls commonly answer with a job_id rather than a result; a cache hit answers inline, with no job.

Report wrapper and upstream version info for scholar-mcp. Returns server_name, server_version, core_version (fastmcp-pvl-core), and (when configured) an upstream version block. Useful for verifying a deployment matches the expected build.