PaperPilot ∢

Research superpowers for AI agents. One MCP server gives any agent instant arXiv + PubMed access with citation-grade extraction.

No scraping. No ads. Wraps the free arXiv and NCBI E-utilities APIs — zero backend risk, zero per-call cost.

Live verification

Tested against both upstream APIs (real responses, this machine):

search_arxiv("CRISPR gene editing", 2)
→ "Primer on the Gene Ontology", Gaudet et al. — https://arxiv.org/pdf/1602.01876v1

search_pubmed("sickle cell CRISPR", 1)
→ JCO Oncol Pract. 2026 Aug 19:OP2600479 — "Translating CRISPR to Practice in the Clinic…"

Install (30 seconds)

Requires Python ≥ 3.10 and mcp>=1.2,<1.10 (1.9.4 verified).

python3.11 -m venv .venv && . .venv/bin/activate
pip install "mcp>=1.2,<1.10"

Add to your Claude / Hermes / Cursor / Codex MCP config:

{
  "mcpServers": {
    "paperpilot": {
      "command": "/path/to/.venv/bin/python",
      "args": ["/path/to/paperpilot/server.py"]
    }
  }
}

Then ask your agent: "Find recent papers on protein folding with diffusion models and cite them." It calls PaperPilot.

Tools

Tool	What it does
search_arxiv(query, max_results=5)	Title, authors, PDF link, summary per paper
search_pubmed(query, max_results=5)	Full citations + abstracts via E-utilities
cite_arxiv(arxiv_id)	Clean citation string + PDF for one arXiv id

Pricing

• Free $0 — 5 calls/day, both sources (get started)
• Pro $99/mo — unlimited calls, BibTeX export, priority cache
• Team $199/mo — 10 agent seats, batch search, white-label API

See landing/index.html for the full sell page. Open a GitHub issue or discussion for Pro/Team provisioning.

Why agent-native?

In the agentic era discovery happens through agents. Whoever the agent reaches for wins. PaperPilot is built to be the tool agents reach for when they need research.