Missed the latest GitHub check; these facts may be older.
Overview
Python MCP server and Claude Code skill that searches papers across arXiv, PubMed, bioRxiv, Semantic Scholar, Crossref, OpenAlex, CORE, Europe PMC and others, then returns standardized, deduplicated results. It downloads PDFs through an open-access fallback chain (source-native, repository discovery, Unpaywall, optional Sci-Hub) and extracts text. Supports stdio, sse and streamable-http transports, with optional OAuth resource-server mode.
Who it is for: Researchers and engineers giving MCP clients access to academic paper search
Strengths
Concurrent multi-source search with deduplication and a single standardized Paper format
Works without API keys; optional keys for Semantic Scholar, OpenAlex, CORE, DOAJ raise limits
Open-access-first download fallback chain, with Sci-Hub off by default
Many install paths: uvx, pip, Docker, Smithery, and a CLI
Weaknesses
Google Scholar search is often blocked by upstream bot detection and rate limits
Several sources are metadata-only or unreliable: PubMed, Crossref, OpenAIRE, BASE, CiteSeerX
Unpaywall is disabled unless an email env var is set
Local HTTP mode without OAuth is open; TLS and rate limits are left to the deployer
What it needs
no GPU
Docker
Needs Python 3.10+, Unpaywall email (optional), API keys for CORE, Semantic Scholar, OpenAlex, DOAJ, IEEE (optional)