cookjohn/zotero-mcp

What it solves

Zotero MCP bridges the gap between AI assistants (such as Claude Desktop, Cursor, or Cherry Studio) and a user's local Zotero reference library. It allows AI models to directly search, read, and manage academic references, annotations, and PDF content, transforming a static reference manager into an active knowledge base for AI-driven literature reviews and research writing.

How it works

The project provides a Zotero plugin with an integrated MCP (Model Context Protocol) server. This server uses the Streamable HTTP protocol to create a real-time bidirectional communication channel between the AI client and the Zotero application. The AI assistant can then call a suite of 20 specialized tools to interact with the Zotero database, including semantic search via embedding vectors (supporting OpenAI and Ollama) and full-text extraction from PDFs.

Who it’s for

Researchers, academics, and students who use Zotero to manage their literature and want to leverage LLMs to analyze their collections, automate citation management, and perform semantic queries across their research papers.

Highlights

• Integrated Architecture: No separate server process is required; the MCP server is built directly into the Zotero plugin.
• Advanced Search: Combines multi-dimensional keyword search, boolean operators, and AI-powered semantic search using embedding vectors.
• Deep Content Access: Extracts full-text from PDFs, notes, abstracts, and webpage snapshots with adjustable detail levels.
• Annotation Analysis: Ability to search and retrieve PDF highlights and annotations filtered by color, tags, or keywords.
• Write Capabilities: Allows AI assistants to create notes, update metadata, manage tags, and import new items via identifiers like DOI or arXiv.
• Privacy-Focused: Operates locally to ensure data privacy.