Literature Review

You can run a multi-source literature review MCP server to search, download, analyze, and generate comprehensive literature reviews across sources. It supports intelligent compression, cross-document synthesis, and Notion export, all orchestrated through MCP endpoints you can call from an MCP client.

How to use

You will use an MCP client to interact with the Literature Review MCP Server. Start by launching the server through a lightweight, remote MCP endpoint or a local process. Once the server is running, you can search across multiple sources, download PDFs in parallel, analyze papers to generate in-depth reviews, and finally export single or cross-document reviews as Markdown or Notion-friendly formats. The workflow is designed to produce detailed analyses (often thousands of words) that compare methods, highlight evolution across papers, and point to future directions.

How to install

Prerequisites You need Node.js (version 18 or newer) and npm to install and run the server. A modern MCP client is required to connect to the MCP server.

# NPX method (recommended)
npx -y @ydzat/literature-review-mcp@latest

# Global install (alternative)
npm install -g @ydzat/literature-review-mcp@latest
literature-review-mcp

Additional sections

Configuration starts with choosing an LLM provider and API key. You can configure the provider via environment variables or a .env file. The available options include SiliconFlow, OpenAI, and Deepseek, with sensible defaults.

Environment variable examples (required and optional):

• LLM_PROVIDER=siliconflow (or openai, custom)
• LLM_API_KEY=your_api_key_here
• LLM_BASE_URL=https://api.siliconflow.cn/v1 (if using a custom endpoint)
• LLM_MODEL=Qwen/Qwen2.5-7B-Instruct (model selection)
• LLM_TEMPERATURE=0.3 (temperature setting)

All data is stored under your home directory for persistence. Typical locations include a database, downloaded PDFs, and generated reviews.

Troubleshooting tips:

• Ensure your API key is valid and the provider URL is reachable.
• If you modify code or configurations, rebuild and restart the server as needed to apply changes.

Available tools

search_academic_papers

Multi-source academic search across DBLP, OpenReview, Papers With Code to surface relevant literature.

batch_download_papers

Parallel download of paper PDFs for bulk analysis.

batch_analyze_papers

Generate individual deep-dive reviews for each paper in bulk.

generate_unified_literature_review

Create a cross-document literature review with ≥4000 words detailing evolution, comparisons, and future directions.

export_individual_review_to_md

Export a single literature review to Markdown format.

batch_export_individual_reviews

Batch export all individual reviews to Markdown.

search_arxiv

Search arXiv for papers.

download_arxiv_pdf

Download PDFs from arXiv.

parse_pdf_to_markdown

Parse PDFs into Chinese Markdown.

convert_to_wechat_article

Convert literature reviews to WeChat article format.

process_arxiv_paper

End-to-end processing of a single arXiv paper.

export_to_notion_full

Export content fully to Notion for collaboration.

export_to_notion_update

Incrementally update Notion with new content.