Loading ToolBench
Resolve DOIs, search ~155M scholarly works, fetch references, and look up publishers via the Crossref REST API. STDIO or Streamable HTTP.
Strong tool definitions with excellent descriptions (avg 250+ chars), comprehensive parameter documentation, and clear pagination patterns. All 5 tools have explicit schemas with typed parameters. Tool names follow verb_noun convention (get_*, search_*). Descriptions are LLM-optimized and include context about when to use each tool and what it returns. Parameter descriptions are detailed with format constraints and examples. Output schemas are documented inline. Minor gaps: no explicit error handling guidance in descriptions, no confirmation patterns for read-only operations, and output field documentation could be more formal.
Resolves a Crossref member ID to its publisher/organization record: primary name, alternate imprint names, owned DOI prefixes, registered DOI counts, a per-work-type breakdown, and per-category metadata deposit coverage. Members are the organizations that register DOIs with Crossref, so this answers "what does this publisher publish, and how completely do they deposit metadata?" Resolve a DOI prefix (e.g. "10.1038") to its member ID with crossref_get_prefix, then pass that ID here.
Resolves a DOI prefix — the registrant portion of a DOI, e.g. "10.1038" — to its owning Crossref member: the publisher name and numeric member ID. Answers "who publishes DOIs starting with 10.1038?" The Crossref prefix record carries only these three facts (no counts, coverage, or flags); the returned memberId chains directly into crossref_get_member for the full publisher record.
Returns the outgoing reference list for a DOI — the works cited by this paper. Each reference includes the deposited citation string and, where Crossref has resolved it, a DOI you can look up with crossref_get_work. Results are paged: referenceCount is the full deposited total, and when more remain the response carries a nextOffset to pass back as offset. Reference list coverage varies by publisher; many older works and non-participating publishers have no indexed references. Incoming citations — the works that cite this paper — are not available through Crossref; use OpenAlex for that.
Resolves a DOI to its full Crossref metadata record: title, authors, affiliations, abstract (when deposited), journal or container, publication date, type, license, full-text links, and funder acknowledgements. The author list is paged: authorCount is the full deposited total, offset and limit select the page (25 authors by default), and when authors remain the response carries a nextOffset to pass back as offset — large-collaboration papers deposit thousands. Outgoing references are reported as a count in referencesCount; the reference entries themselves come from crossref_get_references. The isReferencedByCount field reports the total incoming citation count from Crossref; the citing works themselves are not available through Crossref — use OpenAlex for citation graphs.
Output schemas not formally documented in JSON Schema format. Descriptions mention returned fields (e.g., 'title, authors, affiliations') but no structured schema definition visible in source.
crossref_search_funders has complex parameter interdependencies (works_offset vs works_cursor, query vs funder_doi) documented in description but not formally declared as mutually exclusive in schema.
No explicit error handling guidance in tool descriptions. Descriptions do not explain what happens on invalid DOI format, network failure, or rate limiting, or what the LLM should do next.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | 2026-07-28+ | v2 |
Finds funders registered in the Crossref Funder Registry by name or funder DOI. Provide funder_doi for an exact single-funder lookup — the full DOI ("10.13039/100000001"), the bare registry ID ("100000001"), or either behind a doi: or https://doi.org/ prefix — or query for name-based search. Name-query results page with offset — the nextOffset enrichment carries the value for the following page, up to offset + rows = 100000. Set include_works to true to also return a page of works funded by the matched funder; that list pages two ways. works_offset is the simple one and is capped ten times lower at works_offset + rows = 10000. works_cursor has no ceiling and reaches the whole funded-works list: pass works_cursor="*" on the first call, then chain the nextWorksCursor token from each response. The two cannot be combined, and a cursor walk always starts at the newest work — it cannot resume from an offset. This list also counts works funded by the funder's registry descendants, which a crossref_search_works filter on {"funder": "10.13039/<id>"} does not. Returns funder name, registry ID, country, and alternate names. The Funder Registry supersedes entries, and a deprecated one answers to the same names as its successor while carrying only a fraction of its works: such a record carries replacedBy with the superseding registry ID and the response carries a notice naming it. The replacement is never followed automatically — re-run with funder_doi set to that ID to get the current entry.
DOI parameter accepts multiple formats (bare, resolver-wrapped, doi: prefix) but validation logic not visible in source. LLM cannot verify which formats are actually supported without trial.
Pagination parameters (offset, limit) lack explicit bounds documentation in schema. Descriptions mention ranges (e.g., '1 - 500') but JSON Schema minItems/maxItems not visible.