A Model Context Protocol (MCP) server for scholarly paper search. It gives LLMs and coding agents structured tools to search academic papers, traverse citation graphs, and discover related work across arXiv, Semantic Scholar, OpenReview, OpenAlex, DBLP, Crossref, the ACL Anthology, and the major AI / ML / NLP / CV conference proceedings — all returned as clean, normalized JSON records.

This is the MCP server for resp: the same paper-collection capabilities, exposed as tools any MCP client can call.

• One tool per source. Dedicated tools for arXiv, Semantic Scholar, OpenReview, OpenAlex, DBLP, Crossref, ACM, Connected Papers, and the ACL Anthology.
• Conference-aware search. A single search_conference tool routes to the right source for 27 venues (CVPR, ICCV, ECCV, ICML, NeurIPS, AAAI, EMNLP, IJCAI, and more) — you pass a name and year, it handles the rest.
• Citation graph. Fetch citations, references, and related papers for any paper by id, DOI, or arXiv id.
• Normalized output. Every tool returns the same paper schema (title, authors, year, venue, abstract, doi, pdf_url, num_citations, link, …), so results merge cleanly across sources.
• No keys required. Works out of the box; a free Semantic Scholar key is optional for higher rate limits.
• Lightweight. Pure Python, only requests + beautifulsoup4 + mcp.

• Python 3.9 or newer
• Claude Code, Claude Desktop, Cursor, Windsurf, or any other MCP client

Install the server:

pip install git+https://github.com/monk1337/resp_mcp

This installs a resp-mcp command. Standard MCP config works in most clients:

{
  "mcpServers": {
    "resp": {
      "command": "resp-mcp"
    }
  }
}

Add the server with one command:

claude mcp add resp -- resp-mcp

Or, without installing, run it as a module:

claude mcp add resp -- python -m respmcp.server

Then verify it's connected:

claude mcp list

Add this to your claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "resp": {
      "command": "resp-mcp",
      "env": {
        "SEMANTIC_SCHOLAR_API_KEY": "",
        "RESP_CONTACT_EMAIL": "you@example.com"
      }
    }
  }
}

Restart the app and the Resp tools appear in the tool picker.

Use the same standard config block shown in Getting started — point the command at resp-mcp (or python -m respmcp.server).

The server is configured through environment variables:

To run the server over HTTP instead of stdio:

RESP_MCP_TRANSPORT=http resp-mcp

Then point your MCP client at the HTTP endpoint:

{
  "mcpServers": {
    "resp": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

Supported conferences (search_conference): CVPR, ICCV, WACV, ECCV, ICML, AISTATS, UAI, COLT, ACML, CoLLAs, ACL, EMNLP, EACL, NAACL, CoNLL, COLING, LREC, TACL, Findings, IJCAI, NeurIPS, AAAI, ICPR, KDD, WWW, SIGIR, ICDM.

Once connected, you can ask your agent things like:

• "Search arXiv for recent papers on mixture-of-experts routing."
• "Find EMNLP 2023 papers about retrieval-augmented generation."
• "Get the papers that cite arXiv:1706.03762."
• "Find work related to 'Attention Is All You Need'."
• "Search CVPR 2024 and ECCV 2024 for gaussian splatting papers."

The same capabilities are available as a Python library:

from respmcp import Resp

resp = Resp()  # optional: Resp(semantic_scholar_api_key="...")

papers  = resp.arxiv("multi-label text classification", max_results=10)
citing  = resp.citations("arXiv:1706.03762", max_results=20)
related = resp.related_papers("attention is all you need")
icml    = resp.conference("ICML", "diffusion models", year=2024)

for p in papers:
    print(p.year, p.title, p.link)

git clone https://github.com/monk1337/resp_mcp
cd resp_mcp
pip install -e ".[dev]"
pytest -q            # live integration tests (network required)

MIT