Overview

Zotero MCP is one of the most popular open-source implementations bridging Zotero (the leading reference manager) with the Model Context Protocol (MCP). It enables AI assistants such as Claude Desktop, ChatGPT, Cursor, Cherry Studio, Chorus, and any other MCP-compatible client to directly interact with your personal or group Zotero libraries.

Through the standardized MCP interface, your AI can search papers, retrieve metadata, access full-text PDFs and attachments, extract highlights/notes/annotations, perform semantic searches (in advanced forks), generate summaries, discuss arguments, analyze citation networks—all without leaving the chat interface or manually copying references.

Key Features

Library Browsing & Search: List collections, search by title/author/tag/collection/keyword/full-text
Content Access: Fetch PDF full text, extract highlights, annotations, notes, and comments
Intelligent Interaction: Ask AI to summarize papers, compare findings, explain arguments, or verify claims against your saved literature
Citation & Network Analysis: Explore citation relationships and synthesize insights across multiple papers
Local & Web Modes: Local mode (fast, requires Zotero running) or Web API mode (cloud sync, usually read-only)
MCP Compatibility: Works natively with Claude, ChatGPT connectors, Cursor, and every MCP client
Privacy First: Runs locally via Python; your data never leaves your machine unless using a cloud bridge
Extensible: Community forks add write support (via Zotero plugin), RAG/semantic search, Unpaywall auto-PDF download, etc.

Installation & Quick Start

Install via pip:

pip install zotero-mcp

Set up credentials (Zotero Web API key or enable local API in Zotero preferences)
Launch the server:

zotero-mcp

Add to your AI client configuration (example for Claude Desktop):

{
  "mcpServers": {
    "zotero": {
      "command": "zotero-mcp",
      "env": {
        "ZOTERO_LIBRARY_ID": "your-user-or-group-id",
        "ZOTERO_API_KEY": "your-api-key"
      }
    }
  }
}

Once connected, you can ask questions like:

“Summarize the main arguments in all my papers tagged 'LLM-safety'”
“Find recent papers in my library about retrieval-augmented generation and compare their approaches”

Popular Variants

Several community-maintained forks exist with different trade-offs:

54yyyu/zotero-mcp (most starred) — Clean Python stdio server, local-first, easiest setup
cookjohn/zotero-mcp — Full read/write support when paired with a Zotero plugin
kujenga/zotero-mcp — Solid Web API implementation, great for cloud-synced libraries
mcp-zotero (npm package) — Node.js version optimized for quick remote/cloud access
Hosted bridges (e.g. mcpforzotero.alejandroarnaud.dev) — Zero-install cloud proxy option

Choose based on your needs: local speed, write capability, or no-setup convenience.

Typical Use Cases

Literature Reviews: Let AI synthesize consensus, contradictions, and gaps from your collection
Citation Verification: Cross-check claims against your annotated primary sources
Research Brainstorming: Discuss open questions or potential extensions based on your saved papers
Writing Assistance: Generate cited summaries, outlines, or discussion sections grounded in your library
Automated Research Pipelines: Combine with other MCP servers for multi-tool/agent workflows

Known Limitations

Most implementations are read-only to avoid accidental modifications
Very large libraries (>10,000 items) benefit from forks with semantic/RAG indexing
Local mode requires Zotero application to be running; Web mode needs a valid API key

For the latest forks, configuration examples, and community discussions, visit the main repository and related Zotero/MCP forums.

MIT licensed — completely free for personal, academic, and research use.