Academic Research MCP Server

A unified Model Context Protocol server that provides AI assistants with access to nine academic research APIs through 25 tools, plus local caching, systematic review management, and PRISMA-compliant workflow support. Designed for biomedical researchers who work across clinical medicine and computer science.

APIs & Data Sources

No API keys required for basic use. Optional environment variables for higher throughput:

Installation

Option 1 — One-click (Claude Desktop)

Download academic-research-mcp.dxt and double-click it. Claude Desktop will prompt you for optional API keys and restart automatically.

Option 2 — uvx (Claude Code, Cursor, any MCP client)

No install step — uvx fetches and runs the package in an isolated environment on first use.

Add to your MCP config (.claude/mcp.json, cursor_mcp.json, etc.):

{
  "mcpServers": {
    "academic-research": {
      "command": "uvx",
      "args": ["academic-research-mcp"],
      "env": {
        "OPENALEX_EMAIL": "your-email@institution.edu",
        "S2_API_KEY": "your-key-here",
        "NCBI_API_KEY": "your-ncbi-key-here"
      }
    }
  }
}

Requires uv (curl -LsSf https://astral.sh/uv/install.sh | sh).

Option 3 — pip install

pip install academic-research-mcp
academic-research-mcp  # runs the server

Option 4 — From source

git clone https://github.com/alisoroushmd/academic-research-mcp.git
cd academic-research-mcp
pip install -e .
academic-research-mcp

All 25 Tools

Smart Tools (start here)

• smart_search -- THE recommended search tool. Multi-source search with deduplication and early stopping. Auto-logs to active review when set.
• find_paper -- Universal paper resolver. Accepts any identifier (DOI, PMID, arXiv ID, URL, title).

Unified Search & Authors

• search_papers -- Search any source via source param (openalex, s2, crossref, arxiv, medrxiv, google_scholar, pubmed). For PubMed, supports full MeSH/Boolean syntax. Auto-logs to active review.
• search_authors -- Search for researchers across openalex, s2, orcid, google_scholar.
• get_author -- Detailed author profile by ID (auto-detects source from ID format).
• get_author_works -- Publications by author across any source.
• get_author_funding -- ORCID funding/grants history.

Citation & Recommendation (Semantic Scholar)

• get_paper_network -- Citation network: forward (who cited this), backward (what this cites), or both.
• recommend_papers -- "Papers like this one" recommendations.
• batch_get_papers -- Batch lookup: up to 500 papers in one request.

Preprints (medRxiv/bioRxiv)

• preprints -- Three modes: recent (by category), date_range, or publication status check.

Institutions

• get_institution -- Institution details from OpenAlex.

PDF & Open Access

• open_access -- Find legal PDFs for one or multiple DOIs (Unpaywall, PMC, preprints, repositories).

Citation Validation

• validate_citations -- Verify DOIs and PMIDs actually exist before presenting to users (max 25).

Cache

• cache_manage -- View cache stats or clear by category.

Systematic Review Tools

• create_review -- Start a new systematic review with a name and research question.
• reviews -- List all reviews or get full details (search log, paper counts by status) for one.
• delete_review -- Permanently delete a review and all its data.
• set_active_review -- Activate auto-logging: all subsequent searches auto-deduplicate and log to this review.
• add_papers_to_review -- Manually add papers by DOI/PMID (for expert recommendations, reference lists).
• get_review_papers -- Paginated retrieval with optional status/search filters.
• update_paper_status -- Batch update screening status (new, screened_in, screened_out, included).
• snowball_search -- Citation harvesting: forward/backward from seed papers, deduplicated against the review library.
• export_review -- Export papers as DOI list or full dicts for Zotero import.
• prisma_counts -- PRISMA 2020 flow diagram counts (identification, screening, included).

Systematic Review Workflow

1. create_review("GIM risk SR", "gastric intestinal metaplasia risk stratification")
2. set_active_review(review_id)       # enables auto-logging
3. smart_search / search_papers       # results auto-deduplicate and log
4. snowball_search(review_id, seeds)  # citation harvesting from key papers
5. add_papers_to_review(review_id, ["10.xxxx/..."])  # manual additions
6. get_review_papers(review_id)       # browse candidates
7. update_paper_status(review_id, paper_ids, "screened_in")
8. prisma_counts(review_id)           # PRISMA flow diagram numbers
9. export_review(review_id)           # DOI list for Zotero import

All searches are logged with source, query, filters, raw count, and new-after-dedup count. Deduplication uses DOI (case-insensitive), PMID, exact normalized titles, and >=85% fuzzy title matching only when both the incoming and stored records lack DOI/PMID identifiers.

Abstracts

All search tools return abstracts where available. Semantic Scholar, arXiv, PubMed, and medRxiv return full abstracts directly. OpenAlex reconstructs abstracts from its inverted index format. CrossRef returns abstracts when publishers deposit them (coverage varies). Use abstract content to assess relevance before fetching full papers.

When to Use Which Source

Throughput Strategy

When you need to process papers faster than individual APIs allow:

1. Use smart_search -- automatically picks OpenAlex first (10 req/sec), adds sources as needed
2. S2 batch endpoint -- 500 papers in a single request via batch_get_papers
3. CrossRef fallback -- 50 req/sec with email, 150M+ works
4. PubMed -- 10 req/sec with NCBI API key, full MeSH vocabulary
5. Cache hit -- Repeated lookups (landmark papers, your own work) are instant
6. Google Scholar last -- Most aggressive rate limiting, use sparingly

Snowball harvesting uses a process-shared Semantic Scholar limiter across concurrent harvest calls and both citation endpoints. With S2_API_KEY configured, at most 10 requests may be in flight, while request starts default to one-second spacing to match Semantic Scholar's standard keyed quota across endpoints; if Semantic Scholar grants that key a different quota, set S2_API_KEY_MIN_INTERVAL_SECONDS to the corresponding minimum interval. Without a key, requests are globally serialized with one-second spacing. See the official API tutorial for current quota guidance.

Local Cache

The server includes a SQLite cache (WAL mode, singleton connection, thread-safe) at ~/.cache/academic-research-mcp/cache.db that:

• Caches results from all 9 API clients via @cached decorator
• Avoids redundant API calls for frequently accessed papers
• Default TTL: 24 hours for searches, 7 days for paper details, 3 days for authors
• Expired entries cleaned up automatically on server startup
• Override location with ACADEMIC_CACHE_DIR env var

The same database stores systematic review state (reviews, papers, searches) in separate tables.

Dependencies

• Python 3.10+
• mcp -- Model Context Protocol SDK
• requests -- HTTP client
• httpx[socks] -- async HTTP client with SOCKS proxy support
• scholarly -- Google Scholar access
• defusedxml -- safe XML parsing (PubMed / arXiv responses)
• pip-system-certs -- uses OS certificate store for corporate/VPN environments

License

MIT