Search arXiv for recent papers, retrieve metadata, and download and extract text from arXiv PDFs for reading and analysis.
Search Semantic Scholar for citation-aware paper discovery, retrieve detailed metadata, and align records across sources.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@paper-search-mcp build literature digest on reinforcement learning safety"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
paper-search-mcp is an MCP server for agents that need to search papers, read arXiv PDFs, align records across sources, and produce structured literature-analysis inputs.
The server currently integrates two paper sources:
Semantic Scholar for citation-aware discovery and metadata lookup
arXiv for recent papers, metadata lookup, and PDF text extraction
It also includes higher-level utilities for cross-source alignment, BibTeX export, and compact literature digests.
Chinese documentation is available in README-zh.md.
search_semantic_scholarSearch Semantic Scholar and return normalized paper metadata sorted by citation count.
Parameters:
query: Search query
max_results: Maximum number of results, default 10
get_semantic_scholar_paperFetch detailed metadata for a Semantic Scholar paper by paper_id.
search_arxivSearch arXiv and return normalized metadata.
Parameters:
query: Search query
max_results: Maximum number of results, default 10
sort_by: relevance, lastUpdatedDate, or submittedDate
sort_order: ascending or descending
get_arxiv_paperFetch metadata for one arXiv paper using an arXiv ID, abstract URL, or PDF URL.
read_arxiv_paperDownload an arXiv PDF, cache it locally, extract text from the first pages, and return a structured reading pack.
Parameters:
arxiv_id_or_url: arXiv ID, abstract URL, or PDF URL
max_pages: Maximum number of pages to extract, default 8
max_characters: Maximum number of extracted characters, default 20000
export_bibtexExport a paper as BibTeX.
Parameters:
source: semantic_scholar or arxiv
identifier: Semantic Scholar paper_id or arXiv ID/URL
align_paper_by_titleSearch Semantic Scholar and arXiv by title and return exact normalized title matches across both sources.
Parameters:
title: Paper title used for exact title alignment
semantic_scholar_max_results: Search limit for Semantic Scholar, default 10
arxiv_max_results: Search limit for arXiv, default 10
build_literature_digestSearch across Semantic Scholar and arXiv, deduplicate overlapping papers, and return a compact literature-analysis bundle.
This is useful for downstream agent tasks such as:
finding classic work versus recent work
grouping methods into families
comparing datasets, metrics, and limitations
Related MCP server: Claude Desktop Research MCP Server
This project is designed to use uv for environment and dependency management.
uv syncThis creates .venv in the project directory and installs the project dependencies.
To include development dependencies as well:
uv sync --group devIf you have a Semantic Scholar API key:
export S2_API_KEY=your_key_hereOptional environment variables:
S2_API_KEY: Semantic Scholar API key
PAPER_MCP_HTTP_TIMEOUT: HTTP timeout in seconds, default 30
PAPER_MCP_USER_AGENT: Custom user agent string
PAPER_MCP_CACHE_DIR: Override the on-disk cache directory for downloaded PDFs
For local development or direct Python-based deployment:
pip install .To install directly from a Git repository:
pip install https://github.com/xiaoxiaoxiaotao/paper-search-mcp.gitStart the server directly:
uv run paper-search-mcpExample MCP client configuration:
{
"servers": {
"paper-search": {
"type": "stdio",
"command": "uv",
"args": [
"run",
"--no-sync",
"paper-search-mcp"
],
"cwd": "/home/tao/code/projects/paper-search-mcp",
"env": {
"S2_API_KEY": "${env:S2_API_KEY}"
}
}
},
"inputs": []
}If you prefer not to use input prompts, configure envFile instead of env and place S2_API_KEY=your_key_here in that file.
Semantic Scholar is better for established, citation-rich papers.
arXiv is better for recent work and full-text PDF reading.
build_literature_digest reduces prompt assembly work for downstream agents.
read_arxiv_paper returns text and analysis prompts instead of hard-coded conclusions.
PDF downloads are cached on disk to avoid repeated arXiv fetches.
An npm package is possible as a thin wrapper, but the primary runtime is still Python or Docker.
DOI / PMID / ACL Anthology / OpenAlex support
citation graph and related-paper retrieval
richer section-aware PDF chunking
persistent metadata caching beyond PDFs
Find exact title matches across Semantic Scholar and arXiv and return aligned cross-source pairs.
Inputs: title (required), arxiv_max_results, semantic_scholar_max_results. Returns: object
Search across sources, deduplicate overlapping papers, and return a compact literature review digest.
Inputs: query (required), include_arxiv, max_results_per_source, include_semantic_scholar. Returns: object
Download an arXiv PDF to a specified absolute file or directory path.
Inputs: save_path (required), arxiv_id_or_url (required). Returns: object
Download a Semantic Scholar PDF (Open Access or arXiv linked) to a specified absolute path.
Inputs: paper_id (required), save_path (required). Returns: object
Export a paper as BibTeX using either a Semantic Scholar paper ID or an arXiv identifier/URL.
Inputs: source (required), identifier (required). Returns: object
Fetch normalized metadata for a specific arXiv paper using an arXiv ID, abs URL, or PDF URL.
Inputs: arxiv_id_or_url (required). Returns: object
Fetch detailed metadata for a Semantic Scholar paper by paper ID.
Inputs: paper_id (required). Returns: object
Download an arXiv PDF, extract text from the first pages, and return a reading pack for analysis.
Inputs: arxiv_id_or_url (required), max_pages, max_characters. Returns: object
Search arXiv and return normalized metadata for matching papers.
Inputs: query (required), sort_by, sort_order, max_results. Returns: object
Search Semantic Scholar for papers and return normalized metadata sorted by citation count.
Inputs: query (required), max_results. Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 10 tools
Each tool targets a specific action (search, get, download, align, digest, export) or a specific source (arXiv vs Semantic Scholar), with no overlap in purpose. For example, search_arxiv and search_semantic_scholar clearly differ by source.
All tool names follow a consistent verb_noun pattern using snake_case, with clear action and target (e.g., search_arxiv, get_arxiv_paper, download_arxiv_pdf). The naming is predictable and uniform.
With 10 tools, the surface is well-scoped for a paper search and retrieval domain. Each tool serves a distinct need without being excessive or insufficient.
The tools cover the full workflow: searching both sources, fetching metadata, downloading PDFs, aligning cross-source papers, building digests, and exporting citations. No obvious gaps exist for the stated purpose.
Search arXiv/Semantic Scholar/OpenAlex + medical evidence (PubMed/Europe PMC) + LaTeX/PDF tools.
arXiv and OpenAlex/Crossref paper search (an open Google Scholar alternative) as agent tools.
Academic paper search, scientific literature, citation analysis, arXiv & semantic related-work.
Search arXiv and ACL Anthology, retrieve citations and references, and browse web sources to accel…