Python API

scholar_mcp.api turns each MCP tool into a plain function. MCP, the command line and Python share one implementation, so every function returns a plain dict with the same fields as the tool’s structured output.

from scholar_mcp import api

query = "diffusion policy for dexterous manipulation"
hits = api.search(query, limit=10, year="2024-2026")
for paper in hits["results"]:
    print(paper["year"], paper["title"], paper["id"])

detail = api.info("arXiv:2303.04137", include="detail,citations")
neighbours = api.related("arXiv:2303.04137", relation="similar")
people = api.authors("Fei-Fei Li")
text = api.read("arXiv:2303.04137", pages="1-4")
saved = api.download("10.1038/s41586-021-03819-2", save_dir="papers")
graph = api.graph(["arXiv:2303.04137"], max_hops=1)

Functions

Function	MCP tool	Returns
api.search	search_papers	results, plus a warning when coverage was reduced
api.info	paper_info	paper, citations and references, as requested by include
api.related	recommend_papers	relation, seed, papers and total
api.authors	search_authors	Author profiles
api.read	read_paper	Page-aware text with table and figure selectors
api.download	download_paper	The saved file path and its library collection
api.graph	build_paper_graph	nodes, edges, PageRank, bridges and a Mermaid diagram
api.library	paper_library	The result of the chosen library action

For independent scientific data, api.search_data searches typed objects and api.data_info reads native records. These functions are available through CLI/Python; paper associations use the existing api.info(..., include="data") / paper_info path.

Parameter names and defaults match the MCP tools reference, with one convenience: api.graph accepts seeds as a list as well as a comma-separated string. A lookup that finds nothing returns a dictionary with an error key, the same way the tools report it to a model.

The functions use the same environment variables as the server, so API keys, the data directory and reranker settings from Configuration apply unchanged.