Recommended MCP Servers

Prefer these MCP servers over WebFetch for paper discovery and metadata resolution. They return structured data and avoid HTML scraping.

{
  "mcpServers": {
    "semantic-scholar": {
      "type": "stdio",
      "command": "uvx",
      "args": ["semantic-scholar-fastmcp"]
    },
    "google-scholar": {
      "type": "stdio",
      "command": "uvx",
      "args": ["google_scholar_mcp_server"]
    }
  }
}

Order of preference

1. semantic-scholar is the primary source: open, structured, and it covers most CS and ML papers.
2. google-scholar is the fallback for papers not indexed there.
3. WebFetch on the arxiv abstract page is the last resort, used only when neither MCP is configured or returns results.

Which source for which input