Scholar Feed

What is Scholar Feed?

Scholar Feed is an MCP server that searches over 600,000+ CS/AI/ML research papers with LLM-generated novelty analysis, designed for researchers conducting literature reviews directly in Claude Code, Cursor, or any MCP client.

How to use Scholar Feed?

Run npx scholar-feed-mcp init for an interactive setup wizard, or manually configure your MCP client with the stdio command npx -y scholar-feed-mcp. Anonymous access gives 100 calls/day; set the environment variable SF_API_KEY (get a free key at scholarfeed.org/settings) for 1,000 calls/day. Use tools like search_papers, get_paper, and get_citations to find and explore papers.

Key features of Scholar Feed

• Search 600,000+ arXiv papers indexed daily with multi-signal scoring
• LLM-generated novelty scores from 0.0 to 1.0 for each paper
• Semantic search, citation graph, author discovery, and field orientation tools
• Save papers to a library, create collections, and set up daily watches
• Export BibTeX and fetch full-text LaTeX source
• Support for 11+ MCP clients with config examples provided

Use cases of Scholar Feed

• Technology scouting for recent novel research in a specific area
• Literature review by finding similar papers and exporting BibTeX
• Trend monitoring by checking trending papers in a category
• Author discovery to find top researchers in a field
• Field orientation with an orientation report on a research area

FAQ from Scholar Feed

Do I need an API key?

No, anonymous access works immediately with a limit of 100 calls/day. A free API key from scholarfeed.org/settings raises the limit to 1,000 calls/day.

Which MCP clients are supported?

Claude Code, Cursor, Claude Desktop, Windsurf, Cline / Roo Code, Gemini CLI, LM Studio, JetBrains (PyCharm/IntelliJ), VS Code (GitHub Copilot), Zed, Continue, and any client supporting the standard stdio MCP block.

What is the novelty score?

Every paper includes an llm_novelty_score from 0.0 to 1.0, where 0.7+ indicates a paradigm shift or broad SOTA, 0.5-0.7 novel method with strong results, 0.3-0.5 incremental improvement, and below 0.3 surveys or minor extensions.

What tools are available?

Twenty-five tools covering core search (search_papers, get_paper, get_citations, fetch_fulltext), authors (find_author, co_author_graph), embeddings (embed_text, Pro-only), research (get_field_orientation, get_foundational_lineage), and library/collection/watch/gap analysis tools (require API key).

What are the rate limits?

Per-minute limits range from 10 to 30 per tool. Daily volume is 100 calls anonymously, 1,000 with a free key, and 10,000 on Pro. AI synthesis tools like ask_library have separate limits (1/month free, 200/day Pro).