---
name: paper-search
description: Search, download, and read academic papers through the paper-search MCP tools (mcp__paper-search__*), covering arXiv, PubMed, bioRxiv, Semantic Scholar, Crossref, OpenAlex and 15+ other sources. Use when the user asks to find papers, search research literature, download a paper PDF, or extract text from a paper.
whenToUse: Use when the user asks to find, download, or read academic papers or literature. Prefer these MCP tools over a raw web search for paper discovery.
---

# Paper Search

Academic paper discovery and retrieval in this harness runs through the `mcp__paper-search__*` tools (the paper-search-mcp MCP server). Do not run the `paper-search` CLI or hand-write API requests — use these tools.

## Core tools

- `mcp__paper-search__search_papers` — multi-source concurrent search with deduplication and DOI backfill. Prefer it for broad queries; one call covers many sources at once.
- `mcp__paper-search__download_with_fallback` — OA-first PDF download with a source-native → repository → Unpaywall fallback chain.
- `mcp__paper-search__search_unpaywall` — DOI-centric open-access metadata lookup.

## Source-specific tools

Per source the pattern is `search_<source>` / `download_<source>` / `read_<source>_paper`, with availability varying by source. Sources: arxiv, pubmed, biorxiv, medrxiv, iacr, semantic, crossref, openalex, pmc, core, europepmc, dblp, openaire, citeseerx, doaj, base, zenodo, hal, ssrn, openreview, google_scholar. ACM search is available without an API key. IEEE tools require `PAPER_SEARCH_MCP_IEEE_API_KEY`. Discover the current tool list rather than assuming every source supports full text.

Web of Science (`search_wos`) and Scopus (`search_scopus`, `read_scopus_paper`)
require runtime keys and explicit source selection; default/all searches never
include them. Scopus full-text retrieval requires explicit `full_text=true`;
respect its `full_text`, `abstract_only`, or `unavailable` status. Do not describe
an abstract or metadata record as retrieved full text.

`get_citing_papers` and `get_referenced_papers` provide bounded one-hop OpenAlex
lookups from a DOI or OpenAlex ID. Search caching is disabled by default; enabling
it persists queries and public results and is an operator choice. Use
`get_search_cache_status` to inspect it and `clear_search_cache` only when the
user asks to remove cached results. Clearing does not remove downloaded PDFs.

For speed, prefer a targeted set (`search_arxiv`, `search_semantic`, `search_crossref`) over multi-source search when the field is clear.

## Workflow

1. Search: `search_papers` for broad coverage, `search_<source>` for targeted queries.
2. Present results as a table: title, authors, year, source, DOI/URL.
3. Full text: `read_<source>_paper`; PDF: `download_<source>` or `download_with_fallback`, then report the saved path.

## Notes

- Many public sources work without API keys; optional keys (Semantic Scholar, CORE, Unpaywall email, …) live in `~/.config/paper-search-mcp/.env` and are loaded by the server automatically.
- Google Scholar and SSRN can be bot-blocked upstream; fall back to other sources rather than retrying.