Paper Scout — Data skill for Claude Code

Paper Scout

A personal AI research paper ingestion and scoring pipeline. Paper Scout monitors arXiv, Semantic Scholar, and Hugging Face Daily Papers for new research, scores papers against your interest profile using local embeddings (no LLM calls), and stores candidates in MongoDB for review via MCP tools in Claude Desktop or Cowork.

Two halves:

• Automated pipeline — fetches papers on a daily schedule, scores them, writes candidates to MongoDB. Runs on Fly.io (or anywhere Docker runs).
• MCP server — tools that Claude uses to review candidates, assess significance, manage your interest profile. Runs locally on your machine.

How It Works

Every day at a configured time, Paper Scout:

1. Pulls new papers from arXiv (across your chosen categories), Semantic Scholar trending, and HF Daily Papers
2. Deduplicates against previously seen papers
3. Scores each paper (0.0-1.0) using five weighted signals:

Signal	Default Weight	Source
Embedding similarity	0.45	title+abstract vs your interest profile topics
Keyword match	0.15	keywords found in title/abstract
Citation & author signals	0.20	Semantic Scholar citation count, h-index
Community signal	0.10	HF Daily Papers upvotes
Recency bonus	0.10	Linear decay over fetch window

1. Papers above the candidate threshold get stored as candidates for your review
2. You review them in Claude Desktop / Cowork using the MCP tools

Setup

Prerequisites

• Python 3.12+
• MongoDB (local or Atlas free tier)

1. Clone and install

git clone https://github.com/jonradoff/paper-scout.git
cd paper-scout
python -m venv .venv
source .venv/bin/activate
pip install -e ".[mcp]"

2. Configure

cp .env.example .env

Edit `.env` with your settings:

• MONGODB_URI (required) — your MongoDB connection string. Local mongodb://localhost:27017 works, or use Atlas: `mongodb+srv://