Skip to main content
Literature and prior-art questions come up in research, due diligence and product work alike, and both need an index that is current. Through AgentKey your agent searches Google Scholar and Google Patents results through Serper with the same three calls — find_tools, describe_tool, execute_tool — and can follow a hit into the Academic category for arXiv and AMiner paper records.

What you can ask

  • “Search Google Scholar for ‘protein language model’ papers from the past year.”
  • “Find scholarly articles on graph neural networks for traffic forecasting, second page.”
  • “Which patents mention ‘solid-state battery electrolyte’? List titles and links.”
  • “Are there patents on foldable display hinges that name Samsung?”
  • “Give me 20 Scholar results for ‘LLM evaluation benchmark’ with a Chinese interface.”
  • “Find prior art for a drone package-delivery latch published in the past year.”
  • “Find the arXiv version of this paper and its abstract.”

What’s covered

Scholarly literature
  • Keyword search over Google Scholar results — articles, theses, books and preprints as Scholar indexes them
  • A country code, interface language and free-text location to bias results
  • A time filter — past hour, day, week, month or year — plus result count and page numbers
  • Autocorrect on by default, switchable
Patents
  • Keyword search over Google Patents results
  • The same country, language, location, time-filter, count and paging controls
Beyond Google
  • Paper search and paper detail with normalized records from arXiv and AMiner, plus researcher and institution lookups, live in Academic

How to use it

Pass the request in plain language and let the router pick the tool:
Always describe_tool before executing — recency is the tbs filter, paging is page, and the result fields differ between Scholar and Patents.
Building with code? Scholar results are one of the engines behind the fixed-contract academic.paper.search Scenario Method, which returns a normalized paper page with automatic provider failover; see Scenario Methods. Patent search is available through the catalog tool above.

FAQ

No. Results carry the title, link and snippet as Google returns them. To read a PDF or a patent page, pass the link to a scrape tool; Firecrawl parses PDFs into markdown.
Any Google country code and interface language, plus a free-text location. Queries can be written in any language.
Fetched from Google at request time. The time filter narrows by Google’s own indexing date.