litkit-search

Day-to-day research means bouncing between arXiv, PubMed, Crossref, Semantic Scholar, and a dozen other platforms, hunting for PDFs one by one, and manually checking whether every citation in a manuscript actually exists. litkit-dsh collapses all of that into a single CLI: unified search across 25+ scholarly sources with automatic deduplication, a citation-audit command that verifies references in a docx or pdf manuscript, a multi-stage PDF download chain (open access, institutional proxy, browser-assisted), and one-shot export to BibTeX, RIS, or JSON. Most sources work out of the box with zero API keys; optional keys for Scopus, WoS, and others unlock higher rate limits and richer metadata.

Beyond single commands, the toolkit ships three ready-made workflows: topic-search to fan out around a research theme, deep-search for iterative multi-round retrieval, and bulk-review for batch-processing a collection of papers. A dedicated zh-search command handles Chinese-language literature, and helper commands like doctor and sources make environment troubleshooting and source management straightforward.

On the agent side, litkit-dsh speaks MCP natively, so any MCP-capable client—Claude Desktop, Cursor, Codex, OpenCode, and others—can treat it as a first-class literature tool. A DeepSeek Harness plugin adapter wraps the same functionality for dsh worklets, letting an agent search citations, verify references, and download PDFs inline without leaving its execution context.

It is built for graduate students and researchers writing papers or literature reviews, academic writers who need batch reference management, and developers who want structured scholarly retrieval inside an AI agent pipeline. Released under the MIT license for free use and modification.