paper-search-mcp MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

Unified search_papers across 20+ sources
Platform-specific search + download tools per source
download_with_fallback picks the best source for a DOI
Free-first: public sources by default, paid optional
Python + uvx install

Démo en direct

Aperçu en pratique

paper-search.replay ▶ prêt

0/0

Installer

Choisissez votre client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "paper-search": {
      "command": "uvx",
      "args": [
        "paper-search-mcp"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "paper-search": {
      "command": "uvx",
      "args": [
        "paper-search-mcp"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "paper-search": {
      "command": "uvx",
      "args": [
        "paper-search-mcp"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "paper-search": {
      "command": "uvx",
      "args": [
        "paper-search-mcp"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "paper-search",
      "command": "uvx",
      "args": [
        "paper-search-mcp"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "paper-search": {
      "command": {
        "path": "uvx",
        "args": [
          "paper-search-mcp"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add paper-search -- uvx paper-search-mcp

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : paper-search-mcp

How to build a literature review bibliography from a prompt

👤 Researchers, grad students ⏱ ~30 min intermediate

Quand l'utiliser : New topic, need a starting set of 20-50 papers.

Prérequis

uv — brew install uv

Déroulement

Broad search

search_papers 'mechanistic interpretability attention heads' across arXiv, Semantic Scholar, OpenAlex. Return top 30 by citations.✓ Copié

→ Cross-source list with titles, authors, citation counts
Download the winners

For the top 10 by citation, download_with_fallback to /papers/interp/.✓ Copié

→ PDFs saved
Extract abstracts

Read each PDF's abstract and produce a grouped summary.✓ Copié

→ Grouped bibliography

Résultat : Starter bibliography you can actually read.

Pièges

Duplicate entries across sources — Deduplicate by DOI before ranking
Closed-access papers without OA versions — Use download_with_fallback which tries OA mirrors (arXiv, Unpaywall) first

Combiner avec : filesystem

How to find the latest papers on a topic since a date

👤 Researchers tracking a fast-moving area ⏱ ~20 min beginner

Quand l'utiliser : Weekly/monthly scan.

Déroulement

Date-bounded search

search_arxiv 'constitutional AI' submitted after 2026-01-01. Sort by date desc.✓ Copié

→ Recent list
Read new abstracts

For each new paper, get the abstract and tag: confirms prior work / extends / contradicts / unrelated.✓ Copié

→ Tagged triage

Résultat : Up-to-date view without wading through the whole feed.

Combiner avec : notion

How to cross-check a clinical claim in PubMed + Europe PMC

👤 Clinicians, biotech researchers ⏱ ~25 min intermediate

Quand l'utiliser : A patient-facing doc makes a claim and you want to verify.

Déroulement

Search PubMed

search_pubmed 'metformin cognitive decline' — RCTs only, 2020+.✓ Copié

→ RCT list
Pull full context from Europe PMC if needed

For the top 5, check if Europe PMC has the full text (OA).✓ Copié

→ Full texts where available

Résultat : Source-backed verification.

Pièges

Do not rely on abstracts alone for clinical decisions — Read the actual methodology before citing

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

paper-search + markdownify

Convert downloaded PDFs to markdown for RAG ingestion

For every PDF in /papers/interp/, run pdf-to-markdown and save to /kb/papers/.✓ Copié

paper-search + deep-research

Cross-check a deep research report's sources against arXiv

Given this deep research report, identify any 2024-2026 arXiv papers missing from the citations.✓ Copié

paper-search + notion

Maintain a research reading log

Add each newly-downloaded paper to the Notion 'Reading Log' database with title, authors, DOI, and a 3-line summary.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
search_papers	query, sources?: str[], limit?: int, year_min?: int	Multi-source search	free (public APIs)
download_with_fallback	doi_or_arxiv_id, dest_dir	Get PDF via best available route	free
search_arxiv	query, sort?, year_min?	arXiv-specific	free
download_arxiv	arxiv_id, dest_dir	Get arXiv PDF	free
search_pubmed	query, filters?	Biomedical search	free
search_semantic_scholar	query	Citation-aware search	free
search_biorxiv	query	Biology preprints	free

Coût et limites

Coût d'exécution

Quota d'API: Public APIs have per-IP rate limits (arXiv 3 req/s, PubMed 3/s without key)
Tokens par appel: Search metadata: 200-1000 tokens per paper; PDFs are file-based (no token cost)
Monétaire: Free for public sources; paid sources (IEEE, ACM) need their own keys
Astuce: Get a free NCBI API key for 10x PubMed rate; arXiv has no key but respect the 3 req/s limit.

Sécurité

Permissions, secrets, portée

Portées minimales : Filesystem write to your download directory Optional keys for paid sources

Stockage des identifiants : Env vars for optional API keys

Sortie de données : Hits the public academic APIs you enable

Ne jamais accorder : Nothing destructive — all read/download

Respect per-source terms of use — bulk downloading can get your IP blocked.

Dépannage

Erreurs courantes et correctifs

429 rate limit

Throttle — arXiv needs 3 req/s max, PubMed without key is 3/s. Back off.

download_with_fallback fails for a closed-access DOI

No legal free source exists. Use your library's subscription; this tool only uses OA routes.

Search returns fewer results than the source's website

Some APIs have narrower indexes than their UIs. Cross-check with search_semantic_scholar which has broad coverage.

PDF downloaded but unreadable

Source may return an HTML splash page instead of PDF; check file type and report via the issue tracker.

Vérifier : file path/to/downloaded.pdf

Alternatives

paper-search-mcp vs autres

Alternative	Quand l'utiliser	Compromis
arxiv-mcp	You only need arXiv, nothing else	Single-source only
deep-research MCP	You want synthesis, not just retrieval	LLM costs; different shape

Plus

Ressources

📖 Lire le README officiel sur GitHub

🐙 Voir les issues ouvertes

🔍 Parcourir les 400+ serveurs MCP et Skills