I made a small maintained fork of a Semantic Scholar MCP server. This fork started from JackKuo666/semanticscholar-MCP-Server, which already covered the core idea. My version is mostly a maintenance pass: cleaner packaging, a single Python package entrypoint, better typing and tests, CI, Docker support, optional API-key handling, rate limiting, retries for transient 429s, and paginated citation/reference tools so large papers don’t explode the context window.
The project is simple and just exposes the Semantic Scholar API as Model Context Protocol tools, so an agent can search papers, fetch paper and author details, pull citations and references, and ask for related papers.
The immediate use case was a small literature-mapping pass. I seeded the workflow with a set of papers I already cared about, pulled papers that cited them, and then used a local LLM to categorize the results into my Obsidian vault. That made it much easier to separate directly relevant follow-on work from the broader citation cloud, while keeping the intermediate notes and paper metadata somewhere I could actually revisit.
I should certainly note that there are other Semantic Scholar MCPs. zongmin-yu/semantic-scholar-fastmcp-mcp-server looks like the most visible one and leans into FastMCP plus package-manager installation. smaniches/semantic-scholar-mcp is another compact Python server aimed at Claude Desktop-style use. alperenkocyigit/semantic-scholar-graph-api emphasizes Streamable HTTP transport and broader research-discovery flows. ZidongS/Semantic-Scholar-MCP-Server is another Graph API wrapper in the same general space.
It’s cool that there are so many projects out there on this! Semantic Scholar is an amazing resource and using it with an agent really does feel like an unlock.