Zotero MCP Guide

Zotero MCP Guide

Overview

Zotero MCP is an emerging tool with over 2,000 GitHub stars that implements the Model Context Protocol (MCP) to connect your Zotero reference library with AI assistants such as Claude. By exposing your Zotero library as a structured data source through MCP, it allows AI models to search, retrieve, and reason about your collected academic papers, annotations, and notes during conversations.

The Model Context Protocol provides a standardized way for AI applications to access external data sources. Zotero MCP leverages this protocol to give AI assistants direct, read-only access to your research library. Instead of manually copying and pasting paper details into a chat, you can ask an AI assistant to look up papers in your Zotero library, summarize your annotations on a specific topic, find related references, or help draft text based on your collected sources.

This integration is particularly valuable for researchers who maintain comprehensive Zotero libraries and want to leverage AI assistance for literature reviews, writing, and knowledge synthesis. The AI can work with the full context of your library rather than being limited to whatever you paste into the conversation, making interactions more productive and contextually grounded.

Installation and Setup

Zotero MCP requires both a running Zotero instance and an MCP-compatible AI client.

Prerequisites:

• Zotero 6 or 7 installed and running
• Node.js 18 or later
• An MCP-compatible AI client (Claude Desktop, or another MCP host)

Install the MCP Server:

# Clone the repository
git clone https://github.com/54yyyu/zotero-mcp.git
cd zotero-mcp

# Install dependencies
npm install

# Build the server
npm run build

Configure Zotero:

• Ensure the Zotero local API is enabled (it is by default on port 23119)
• The MCP server communicates with Zotero through its local HTTP API
• No API key is needed for local connections; Zotero must simply be running

Configure Your AI Client:

For Claude Desktop, add the Zotero MCP server to your configuration file (claude_desktop_config.json):

{
  "mcpServers": {
    "zotero": {
      "command": "node",
      "args": ["/path/to/zotero-mcp/dist/index.js"]
    }
  }
}

For other MCP hosts, follow their respective documentation for adding MCP servers. The server communicates over stdio using the standard MCP transport protocol.

Verify the Connection:

1. Start Zotero
2. Launch your AI client with the MCP configuration
3. Ask the AI to search your Zotero library (e.g., "Find papers about machine learning in my Zotero library")
4. If the AI returns results from your library, the connection is working

Core Features

Library Search: The MCP server exposes a search tool that allows AI assistants to query your Zotero library using natural language or structured queries. The AI can search by:

• Title keywords
• Author names
• Publication year or date range
• Tags and collections
• Full-text content (if Zotero full-text indexing is enabled)

Item Retrieval: Retrieve complete metadata for specific items including:

• Bibliographic information (title, authors, journal, year, DOI, etc.)
• Abstract and keywords
• Tags and collection membership
• Attached file information
• Related items

Annotation Access: The AI can read your PDF annotations and notes for specific items. This means you can ask questions like "What did I highlight in the Smith 2024 paper?" or "Summarize my annotations on papers about transformer architectures" and the AI will access your actual annotations to formulate a response.

Collection Browsing: Navigate your Zotero collection hierarchy to understand how you have organized your library. The AI can list collections, show items within specific collections, and understand the taxonomic structure of your research.

Note Reading: Access standalone notes and item notes in your Zotero library. This is useful for AI-assisted writing where your accumulated research notes provide the source material for drafting text.

Citation Formatting: Request formatted citations for items in your library. The AI can generate citations in various styles (APA, MLA, Chicago, etc.) by accessing the complete bibliographic data through MCP.