Papers With Code MCP

Papers With Code MCP plugs Claude directly into the Papers With Code API, letting you search for machine learning papers, datasets, benchmark results, and author profiles without leaving your editor. It's a narrow tool with a clear mandate: if you're doing ML research or due diligence on state-of-the-art methods, it saves you from tabbing out to the browser every five minutes. I'd reach for this when I'm trying to figure out which dataset to use for a new project, or when I need to check if a paper's claims hold up against recent benchmarks. The search is fast and returns structured data—paper titles, abstracts, GitHub links, dataset descriptions, leaderboard positions. You can ask Claude to find papers on a topic, then immediately pull down the associated code repo or dataset without manual copy-paste. The main quirk is that it's only as good as Papers With Code's index. If a paper isn't on the platform, you won't find it here. That's fine for mainstream ML research, but if you're working in a niche subfield or need older pre-arXiv papers, you'll still need Google Scholar or Semantic Scholar. The tool also doesn't do full-text search inside papers—it's metadata and abstracts only. Setup is trivial if you're on Claude Desktop: one npx command and it's live. No API keys, no environment variables, no config file wrangling. That's rare enough to be worth mentioning. The repo is minimal—just the MCP server implementation—so don't expect a lot of hand-holding or advanced features. It does one thing and does it cleanly. Who shouldn't bother: if you're not actively doing ML research or building models, this is too specialised. It won't help you write better code or debug faster. It's for people who need to know what SOTA looks like on ImageNet or whether a dataset has a test split, not for general-purpose development.