papernexus-mcp

MCP server for academic paper searching, fetching, and reference exploration.

Tools

| Tool | Description | | ------ | ------------- | | paper_searching | Enrich Google Scholar results with metadata from Semantic Scholar, arXiv, Unpaywall | | paper_fetching | Fetch full paper as markdown (cache-first, arxiv2md or MinerU PDF conversion) | | paper_references | Extract cited references from paper markdown, enrich each with metadata | | paper_reading | AI-powered paper reader (not yet implemented) |

Setup

npm install
cp .env.example .env
# Fill in your API keys in .env

Usage

npm run mcp          # Start MCP server (stdio transport)
npm test             # Run tests
npm run build        # Compile TypeScript

Environment Variables

| Variable | Purpose | Required | | ---------- | --------- | ---------- | | TOKEN_MINERU | MinerU PDF OCR API | Yes | | EMAIL_UNPAYWALL | Unpaywall OA lookup | Yes | | DIR_CACHE | Cache directory (default: .cache/) | No |

Architecture

apify MCP (external) → paper_searching → paper_fetching → paper_reading
                                                        ↗
                                          paper_references → loop

paper_searching does NOT call apify — apify is invoked at the skill/orchestration layer
paper_fetching does NOT search — it only accepts PaperMeta with existing URLs
paper_references reuses paper_searching logic at code level

License

Apache License 2.0

MCP Servers