Zotero MCP Server: Bringing Your Research Library to popular AI Chatbots

Update (September 22, 2026): the configuration below is updated for Zotero MCP Server 4.0 — JavaScript and Python shells over a resident C++ core, with no need to open Zotero. For the full feature overview, benchmark numbers and FAQ, read the Zotero MCP Server 4.0 guide.

In 2026, the way we interact with AI has changed. We are no longer limited to the chat box on a single website. With the rise of the Model Context Protocol (MCP), AI agents can now "see" and "interact" with your local data.

For Zotero users, this means something revolutionary: Your entire research library can now be the brain of your favorite AI chatbot.

What is the Zotero MCP Server?

The Zotero MCP Server (v4.0.1) is a bridge that allows external AI clients (like Claude Code, ChatWise, Cherry Studio, Cursor, Codex and the Gemini CLI) to query your Zotero library directly.

Instead of copying and pasting text into a chat window, you can simply ask the AI, "What does my Zotero library say about renewable energy trends?" and the AI will use the MCP server to search your papers, retrieve the relevant text, and answer you with citations.

Core Features of the PapersGPT MCP Server

1. Semantic & Metadata Search: Built on the powerful docsagent engine, the server performs BM25 full-text searches across your titles, creators, tags, abstracts, and even your personal annotations.
2. Instant Full-Text Retrieval: AI agents can "reach in" and grab the specific text of a PDF in your Zotero library using its primary key.
3. Blazing Fast Performance: The indexing and retrieval engine is a resident C++ search core, so responses stay in the millisecond range — and you can drive it from either a JavaScript shell (npx) or a Python shell (pip), whichever your environment already has.
4. Local-First Privacy: Your Zotero database stays on your machine. The MCP server only provides context to the AI when you specifically ask it to.

How to Set It Up in 60 Seconds

Zotero MCP Server 4.0 runs a background search core. Start it once, then register the server in whichever MCP client you use — the JavaScript and Python shells are interchangeable, and Zotero does not need to be open for search:

JavaScript / Node (npm)

npx @docsagent/mcp-zotero start

{
  "mcpServers": {
    "docsagent-zotero": {
      "command": "npx",
      "args": [
        "-y",
        "@docsagent/mcp-zotero"
      ]
    }
  }
}

Python (PyPI)

pip install docsagent-mcp-zotero
docsagent-mcp-zotero core start

{
  "mcpServers": {
    "docsagent-zotero": {
      "command": "docsagent-mcp-zotero",
      "args": []
    }
  }
}

Note: both shells read the same config file at ~/.docsagent/config.json. If your Zotero data directory is not the default ~/Zotero, set zoteroDataDir there (or point DOCSAGENT_CONFIG at your own file).

Use Cases: Why Researchers Love MCP

• In-Editor Research: Using Cursor to write your paper? Connect the Zotero MCP server to cite your own library as you type.
• Cross-Library Synthesis: Ask a "thinking" model (like GPT-5 or Gemini 2.0 Thinking) to find connections between papers in your Zotero library and the latest web search results.
• Automated Literature Discovery: Use an AI agent to monitor your library and automatically tag new papers based on their conceptual similarity to your existing collections.

Conclusion

The Zotero MCP Server is more than just a plugin; it's a fundamental shift in research productivity. By opening up your Zotero library to the broader AI ecosystem, PapersGPT ensures that your hard-earned research is always at your fingertips, no matter which AI tool you prefer to use.

Ready to connect your library? Check out the Zotero MCP documentation to get started.