Academic Research MCP — Paper Search, DOIs & Citations (arXiv + Crossref + OpenAlex)

An MCP server for AI agents — and a normal Apify scraper — that turns scholarly metadata into agent-callable tools. Point Claude, Cursor, or any MCP client at it and your agent can run a full literature review: search published papers and the latest preprints, look up a paper by DOI, and trace citations both directions — across Crossref (160M+ works), OpenAlex (240M+ works), and arXiv.

Built for the AI-agent era: clean JSON in, structured results out, no API keys to manage, no HTML scraping.

Why this one

Most research tools on the store hit a single source and only go one direction. This one aggregates three stable scholarly APIs behind a consistent tool set and gives an agent the forward citation graph (who cites a paper), not just its bibliography — so it can ask "what's the newest work building on this?"

• Zero selector rot. Wraps first-party JSON/XML APIs, not scraped HTML — it stays working.
• No keys, no logins. Crossref (CC-BY), OpenAlex (CC0), and arXiv are public, reuse-friendly APIs.
• Fair billing. You're only charged when a tool returns a successful result.

Tools (MCP mode)

Tool	What it does
search_papers	Search peer-reviewed papers across Crossref, OpenAlex, or all (merged + de-duplicated, ranked by citations).
search_preprints	Search the latest arXiv preprints; optional category filter (cs.AI, stat.ML, …).
get_paper_by_doi	Full metadata for one paper by DOI.
get_references	The works a paper cites (its bibliography) — backward tracing.
get_citations	The works that cite a paper (forward citation graph), most-cited first.
get_author_works	An author's most-cited works + metrics (works count, total citations, h-index).

Connect (MCP client)

With the actor running in Standby mode, connect your MCP client to:

https://<your-username>--academic-research-mcp.apify.actor/mcp

Use the streamable HTTP transport with an Authorization: Bearer <APIFY_TOKEN> header:

{
  "mcpServers": {
    "academic-research": {
      "url": "https://<your-username>--academic-research-mcp.apify.actor/mcp",
      "headers": { "Authorization": "Bearer <APIFY_TOKEN>" }
    }
  }
}

Then ask your agent things like "Find the most-cited papers on retrieval-augmented generation since 2024, then show me what's cited the top one this year."

Use it as a normal scraper (no MCP)

You can also just Run it with input and get a dataset back:

{ "operation": "search_papers", "query": "diffusion models", "source": "all", "fromYear": 2024, "maxResults": 15 }

Other operations: search_preprints (+ category), get_paper_by_doi / get_references / get_citations (+ doi), get_author_works (+ author).

Pricing

Pay-per-event: $0.004 per successful MCP tool call (failed/empty calls are free), or $0.00001 per result in a normal dataset run. A typical 30-call literature-review session costs well under $0.20.

Data sources & terms

• Crossref REST API — public, no auth, metadata under CC-BY.
• OpenAlex API — public, no auth, CC0 data.
• arXiv API — public, no auth; results subject to arXiv's terms.

Public, logged-out data only. No login-gated content, no anti-bot circumvention. Not a medical/clinical data source.

Local development

npm install
npm run build
# MCP / Standby mode:
APIFY_META_ORIGIN=STANDBY ACTOR_WEB_SERVER_PORT=8080 npm run dev
# Standard run (writes to local dataset):
node dist/main.js

---

Built with Claude Code. Wraps public scholarly APIs; not affiliated with Crossref, OpenAlex, or arXiv.