research-papers-mcp 1.0.0

MCP server for research paper discovery, citation analysis, and trend detection across arXiv, PubMed, and Semantic Scholar

research-papers-mcp

Research paper intelligence for LLMs

Search, analyze, and track 200M+ papers from arXiv, PubMed, and Semantic Scholar -- right from Claude, Cursor, or any MCP client.

Live demo: https://huggingface.co/spaces/barissozudogru/research-papers-mcp

---

• Federated Search -- Query arXiv, PubMed, and Semantic Scholar in a single call
• Self-Growing Corpus -- Local SQLite cache grows with every search, enabling richer analysis over time
• Zero Config -- No database servers, no API keys required, no background workers
• 10 Research Tools -- From paper discovery to real citation graphs to BibTeX export

Quick Start

pip install research-papers-mcp

Add to your MCP client config (Claude Desktop, Claude Code, Cursor, etc.):

{
  "mcpServers": {
    "research-papers": {
      "command": "research-papers-mcp",
      "args": []
    }
  }
}

That's it. Start asking your AI about research papers.

[!TIP] No API keys are required to get started. All three sources work without authentication. A Semantic Scholar API key is optional and only needed for higher rate limits.

See It in Action

You: Find recent papers on transformer efficiency and model compression

Claude (using search_papers): Found 34 papers across arXiv and Semantic Scholar. 12 new papers cached. Here are the top results by impact score...

You: What's trending in this area?

Claude (using get_trending_topics): Based on your cached corpus of 847 papers:

• knowledge distillation -- emerging (+94%)
• model pruning -- stable
• quantization -- emerging (+67%)

You: Find papers similar to the knowledge distillation one

Claude (using find_similar_papers): Found 8 similar papers via SPECTER2 embeddings:

1. "DistilBERT, a distilled version of BERT"
2. "TinyBERT: Distilling BERT for NLU" ...

You: Who cites the original knowledge distillation paper?

Claude (using get_paper_citations): Found 1,247 citations (89 influential). Top citing papers include...

You: Export those as BibTeX

Claude (using export_bibtex): Here are 12 BibTeX entries ready for your LaTeX document.

Tools

Discovery

Tool	Description
search_papers	Federated search across arXiv, PubMed, and Semantic Scholar
search_cached_papers	Fast local search with date, source, and citation filters
get_paper_details	Full metadata for any cached paper

Analysis

Tool	Description
get_paper_citations	Real citation and reference data from Semantic Scholar
find_similar_papers	SPECTER2 semantic similarity with TF-IDF fallback
get_author_profile	Publication frequency, top topics, and collaborators

Intelligence

Tool	Description
get_trending_topics	Detect emerging and declining research topics over time
generate_literature_review	Structured review with subtopics and consensus analysis
export_bibtex	Export papers as BibTeX for LaTeX documents
get_cache_stats	Cache size, source breakdown, and date range

Resources

URI	Description
papers://stats	Database statistics (total papers, by source, date range)
papers://fields	Research fields in the corpus with paper counts

How It Works

                    search_papers
                         |
          +--------------+--------------+
          |              |              |
       arXiv          PubMed    Semantic Scholar
          |              |              |
          +--------------+--------------+
                         |
                    SQLite Cache
                  (grows over time)
                         |
          +--------------+--------------+
          |              |              |
      Trends       Similarity      Citations
      Detection    (SPECTER2)     (S2 API)

1. Search -- search_papers fetches from external APIs and caches results locally
2. Analyze -- Citation data from Semantic Scholar API, similarity via SPECTER2 embeddings (with TF-IDF fallback), trend detection on cached corpus
3. Grow -- The cache accumulates with every search, enabling richer trend detection and analysis

No PostgreSQL, Redis, or background workers. Everything runs in a single process with zero-config SQLite.

Transport options

# stdio (default -- for MCP clients like Claude Desktop)
research-papers-mcp

# SSE (for web-based MCP clients)
research-papers-mcp --transport sse --port 8080

# Streamable HTTP (for team/remote deployment)
research-papers-mcp --transport streamable-http --port 8080

Configuration

Variable	Description	Required
SEMANTIC_SCHOLAR_API_KEY	Higher rate limits for Semantic Scholar API	No
RESEARCH_MCP_CACHE_DIR	Custom cache directory (default: ~/.research-papers-mcp/)	No

Development

git clone https://github.com/barissozudogru/deep-research-digest.git
cd deep-research-digest
pip install -e ".[dev]"
pytest

See CONTRIBUTING.md for guidelines.

License

MIT