kivv
Automated arXiv research paper discovery and AI-powered summarization system that enables Claude Desktop users to search, retrieve, and get intelligent summaries of academic papers through MCP integration.
README
<picture> <source media="(prefers-color-scheme: dark)" srcset="assets/hero-dark.svg"> <source media="(prefers-color-scheme: light)" srcset="assets/hero.svg"> <img alt="kivv - arXiv Research Assistant" src="assets/hero.svg"> </picture>
Automated arXiv research paper discovery and AI-powered summarization system with MCP (Model Context Protocol) integration for Claude Desktop. Built as a vibe-coding project, we keep the tooling lightweight and the feedback loop fast so you can stay in flow while shipping research insights.
Features
- Daily Automation: Automatically searches arXiv for papers matching your topics
- AI Summaries: Claude-powered intelligent paper summarization with cost optimization
- Multi-User: Support for multiple users with independent topic configurations
- MCP Integration: Direct integration with Claude Desktop via Model Context Protocol
- RSS Feeds: Per-user RSS/Atom feeds for any feed reader
- Web Dashboard: Optional SvelteKit dashboard (coming soon)
- Cost-Effective: Runs mostly on Cloudflare free tier (~$3/month for 2 users)
Architecture
- MCP Server: TypeScript Worker handling MCP protocol and tool execution
- Daily Automation: Cron-triggered Worker for paper collection and summarization
- Storage: Cloudflare D1 (SQLite), R2 (PDFs), KV (cache)
- AI: Claude 3.5 Sonnet with Haiku triage for cost optimization
Quick Start
Prerequisites
- Cloudflare account with Workers, D1, KV, and R2 enabled
- Anthropic Claude API key (console.anthropic.com)
- Bun 1.1+ (package manager - install from bun.sh)
- Wrangler CLI (installed automatically via bun)
Installation
# Clone the repository
git clone https://github.com/jeffaf/kivv.git
cd kivv
# Install dependencies with bun (our vibe-coding default)
bun install
# Set up environment variables
cp .env.example .env
# Edit .env with your API keys and Cloudflare credentials
Deployment
For complete deployment instructions, see DEPLOYMENT.md.
Quick deployment:
# Set secrets for automation worker
cd automation
wrangler secret put CLAUDE_API_KEY
wrangler secret put CRON_SECRET
# Deploy automation worker
wrangler deploy
# Deploy MCP server
cd ../mcp-server
wrangler deploy
# Configure Claude Desktop (see DEPLOYMENT.md for details)
# Add MCP server URL and API key to Claude Desktop config
Verification
# Test automation worker
curl https://kivv-automation.<username>.workers.dev/health
# Test MCP server
curl https://kivv-mcp.<username>.workers.dev/health
# Check database
wrangler d1 execute kivv-db --command "SELECT COUNT(*) FROM users"
# Or use the automated health check script
./scripts/health-check.sh
See TROUBLESHOOTING.md if you encounter any issues.
Automated Deployment Script
For one-command deployment:
# Run the automated deployment script
./scripts/deploy.sh
# This will:
# - Verify prerequisites
# - Check infrastructure
# - Configure secrets
# - Deploy both workers
# - Provide Claude Desktop config
Project Structure
kivv/
├── mcp-server/ # MCP Server Worker (Claude integration)
├── automation/ # Daily automation Worker (cron)
├── shared/ # Shared types and utilities
├── tests/
│ ├── security/ # Security tests (auth, injection, XSS)
│ ├── integration/ # MCP tool integration tests
│ └── unit/ # Unit tests
├── .checkpoint/ # Development checkpoints
└── package.json # Monorepo root with workspaces
Documentation
- Deployment Guide - Complete production deployment instructions
- Troubleshooting Guide - Common issues and solutions
- Setup Checklist - Infrastructure setup
- Implementation Plan - Chunked development guide
- API Documentation - API endpoints and usage
- PRD (Full Spec) - Complete technical specification
CI & Automation
- Tests & Type Checks: GitHub Actions run
bun run type-checkandbun teston pushes and pull requests to keep the vibe coding fast without breaking the build. - Deployments: Cloudflare Workers deploy pipelines trigger when
mcp-server/orautomation/change, usingwranglerwith your configured Cloudflare credentials.
Development
# Run type checking
bun run type-check
# Run all tests
bun test
# Run security tests specifically
bun run test:security
# Run tests in watch mode
bun run test:watch
# Run MCP server locally
bun run dev:mcp
# Run automation worker locally
bun run dev:automation
# Build all workspaces
bun run build
Testing
The project includes comprehensive test coverage:
-
Security Tests (100% coverage required):
- Authentication (API key validation)
- Authorization (user data isolation)
- SQL injection prevention
- XSS prevention in RSS feeds
- Rate limiting enforcement
-
Integration Tests: End-to-end MCP tool workflows
-
Unit Tests: Isolated utility function testing
# Run with coverage report
bun run test:coverage
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。