Paper Search

paper_search/
├── models/          # Pydantic 数据模型 (Paper, SearchResult, SnowballResult)
├── connectors/      # 平台连接器 (@register 自动注册)
│   ├── base.py      #   抽象基类 PaperConnector
│   ├── registry.py  #   连接器注册表（自动发现）
│   ├── arxiv.py     #   arXiv 连接器
│   ├── semantic.py  #   Semantic Scholar 连接器
│   └── ...          #   25+ 连接器
├── service/         # 业务编排层（不依赖任何传输协议）
│   ├── search_service.py    # 多源搜索 + Snowball
│   ├── download_service.py  # 下载回退链
│   └── export_service.py    # CSV/RIS/BibTeX 导出
├── transports/      # 传输适配层
│   └── mcp_server.py        # MCP Server（薄包装）
├── config.py        # 环境变量管理
└── utils.py         # 工具函数

# 克隆项目
git clone https://github.com/its-antony/paper-search.git
cd paper-search

# 安装依赖（自动创建虚拟环境）
uv sync

# 如果需要 MCP Server 功能
uv sync --extra mcp

import asyncio
from paper_search import PaperSearchService

async def main():
    svc = PaperSearchService()

    # 多源并发搜索
    result = await svc.search(
        "real world assets tokenization",
        sources="crossref,openalex,arxiv",
        max_results_per_source=5,
    )

    for paper in result.papers:
        print(f"{paper.title} ({paper.source})")
        print(f"  DOI: {paper.doi}")

asyncio.run(main())

async def snowball_demo():
    svc = PaperSearchService()

    result = await svc.snowball(
        paper_id="DOI:10.1038/nature12373",
        direction="both",      # forward + backward
        depth=2,               # 2 层递归
        max_results_per_direction=10,
    )
    print(f"发现 {result.total} 篇相关论文")

from paper_search.service import DownloadService, ExportService
from paper_search.connectors.registry import ConnectorRegistry

async def download_and_export():
    reg = ConnectorRegistry()
    dl = DownloadService(registry=reg)
    export = ExportService()

    # 下载论文 PDF（默认只走 open-access fallback）
    path = await dl.download(
        source="arxiv",
        paper_id="2106.12345",
        save_path="./downloads",
    )

    # 导出搜索结果
    svc = PaperSearchService()
    result = await svc.search("blockchain finance", sources="crossref")
    export.export(result.papers, format="bibtex", save_path="./exports")

{
  "mcpServers": {
    "paper-search": {
      "command": "uv",
      "args": [
        "run", "--directory", "/path/to/paper-search",
        "-m", "paper_search.transports.mcp_server"
      ],
      "env": {
        "PAPER_SEARCH_MCP_UNPAYWALL_EMAIL": "your@email.com",
        "PAPER_SEARCH_MCP_SEMANTIC_SCHOLAR_API_KEY": ""
      }
    }
  }
}

# 添加为全局 MCP Server（所有项目可用）
claude mcp add --scope user paper-search -- \
  uv run --directory /path/to/paper-search -m paper_search.transports.mcp_server

# 或仅当前项目可用
claude mcp add paper-search -- \
  uv run --directory /path/to/paper-search -m paper_search.transports.mcp_server

# 如需配置 API Key
claude mcp add --scope user paper-search \
  -e PAPER_SEARCH_MCP_UNPAYWALL_EMAIL=your@email.com \
  -e PAPER_SEARCH_MCP_SEMANTIC_SCHOLAR_API_KEY=sk-xxx \
  -- uv run --directory /path/to/paper-search -m paper_search.transports.mcp_server

# 启动 HTTP 模式（局域网可访问）
uv run -m paper_search.transports.mcp_server --transport streamable-http --host 0.0.0.0 --port 8000

# 或 SSE 模式
uv run -m paper_search.transports.mcp_server --transport sse --host 0.0.0.0 --port 8000

claude mcp add --scope user paper-search --transport http http://192.168.x.x:8000/mcp

cp .env.example .env
# 编辑 .env 填入你的 key

# paper_search/connectors/my_platform.py
from ..models.paper import Paper
from .base import PaperConnector, ConnectorCapabilities
from .registry import register

@register("my_platform")
class MyPlatformConnector(PaperConnector):
    capabilities = ConnectorCapabilities(search=True, download=True, read=False)

    def search(self, query: str, max_results: int = 10, **kwargs):
        # 你的搜索逻辑
        return [Paper(paper_id="...", title="...", source="my_platform")]

    def download_pdf(self, paper_id: str, save_path: str) -> str:
        # 你的下载逻辑
        return f"{save_path}/{paper_id}.pdf"

# 安装开发依赖
uv sync --extra dev --extra mcp

# 运行测试
uv run pytest tests/ -v

# 启动 MCP Server（stdio 模式）
uv run paper-search-mcp

# 启动 MCP Server（HTTP 模式，局域网可访问）
uv run paper-search-mcp --transport streamable-http --host 0.0.0.0 --port 8000