Skip to main content

ArXivIntegration

What it is​

  • An integration wrapper around the arxiv Python client for:
    • Searching ArXiv papers
    • Fetching metadata for a specific ArXiv paper
  • Can also expose these capabilities as LangChain StructuredTool tools.

Public API​

ArXivIntegrationConfiguration​

  • Dataclass extending IntegrationConfiguration.
  • Fields:
    • max_results: int = 10 — default maximum number of search results when not provided per-call.

ArXivIntegration​

Integration class (extends naas_abi_core.integration.Integration).

__init__(configuration: ArXivIntegrationConfiguration)​

  • Creates an arxiv.Client() and stores configuration.

search_papers(query: str, max_results: Optional[int] = None) -> List[dict]​

  • Searches ArXiv with arxiv.Search(query=..., max_results=...).
  • Returns a list of paper metadata dictionaries with keys:
    • id, title, authors, summary, published, updated, categories, links, pdf_url

get_paper(paper_id: str) -> dict​

  • Fetches metadata for one paper using arxiv.Search(id_list=[paper_id]).
  • Returns a metadata dictionary with the same keys as search_papers.

as_tools(configuration: ArXivIntegrationConfiguration) -> List[StructuredTool] (staticmethod)​

  • Builds and returns two LangChain StructuredTool instances:
    • search_arxiv_papers → calls search_papers
    • get_arxiv_paper → calls get_paper
  • Uses Pydantic argument schemas defined inside the method.

Configuration/Dependencies​

  • Depends on:
    • arxiv (Python package)
    • langchain_core.tools.StructuredTool
    • pydantic (BaseModel, Field)
    • naas_abi_core.integration (Integration, IntegrationConfiguration)
  • Configuration:
    • ArXivIntegrationConfiguration.max_results controls default search result count.

Usage​

Direct integration usage​

from naas_abi_marketplace.applications.arxiv.integrations.ArXivIntegration import (
    ArXivIntegration,
    ArXivIntegrationConfiguration,
)

cfg = ArXivIntegrationConfiguration(max_results=5)
client = ArXivIntegration(cfg)

papers = client.search_papers("cat:cs.CL")
print(papers[0]["id"], papers[0]["title"])

paper = client.get_paper(papers[0]["id"])
print(paper["pdf_url"])

As LangChain tools​

from naas_abi_marketplace.applications.arxiv.integrations.ArXivIntegration import (
    ArXivIntegration,
    ArXivIntegrationConfiguration,
)

tools = ArXivIntegration.as_tools(ArXivIntegrationConfiguration(max_results=3))

result = tools[0].func(query="quantum computing", max_results=2)
print(len(result))

Caveats​

  • get_paper uses next(self.__client.results(search)); if no results are returned for the given paper_id, it will raise StopIteration.
  • id is derived from paper.entry_id.split("/")[-1] (the last URL segment of the entry id).