Zotero MCP MCP Server

Model Context Protocol server for Zotero

<!-- Ownership verification marker for the official MCP registry, which matches it against the name in server.json when this README is rendered on PyPI: https://modelcontextprotocol.io/registry/package-types --> <!-- mcp-name: io.github.kujenga/zotero-mcp -->

This project is a python server that implements the Model Context Protocol (MCP) for Zotero, giving you access to your Zotero library within AI assistants. It is intended to implement a small but maximally useful set of interactions with Zotero for use with MCP clients.

<a href="https://glama.ai/mcp/servers/jknz38ntu4"> <img width="380" height="200" src="https://glama.ai/mcp/servers/jknz38ntu4/badge" alt="Zotero Server MCP server" /> </a>

Features

This MCP server provides the following tools:

• zotero_search_items: Search for items in your Zotero library using a text query
• zotero_item_metadata: Get the complete metadata for a specific Zotero item, covering every populated field grouped into publication details, identifiers, timestamps, and library information
• zotero_item_fulltext: Get the full text of a specific Zotero item (i.e. PDF contents)

These can be discovered and accessed through any MCP client or through the MCP Inspector.

Each tool returns formatted text containing relevant information from your Zotero items, and AI assistants such as Claude can use them sequentially, searching for items then retrieving their metadata or text content.

Installation

This server can either run against either a local API offered by the Zotero desktop application) or through the Zotero Web API. The local API can be a bit more responsive, but requires that the Zotero app be running on the same computer with the API enabled. To enable the local API, do the following steps:

1. Open Zotero and open "Zotero Settings"
2. Under the "Advanced" tab, check the box that says "Allow other applications on this computer to communicate with Zotero".

To use the Zotero Web API, you'll need to create an API key and find your Library ID (usually your User ID) in your Zotero account settings here: https://www.zotero.org/settings/keys

These are the available configuration options:

• ZOTERO_LOCAL=true: Use the local Zotero API (default: false, see note below)
• ZOTERO_API_KEY: Your Zotero API key (not required for the local API)
• ZOTERO_LIBRARY_ID: Your Zotero library ID (your user ID for user libraries, not required for the local API)
• ZOTERO_LIBRARY_TYPE: The type of library (user or group, default: user)

uvx with Local Zotero API

To use this with Claude Desktop and a direct python install with uvx, add the following to the mcpServers configuration:

{
  "mcpServers": {
    "zotero": {
      "command": "uvx",
      "args": ["zotero-mcp@latest"],
      "env": {
        "ZOTERO_LOCAL": "true",
        "ZOTERO_API_KEY": "",
        "ZOTERO_LIBRARY_ID": ""
      }
    }
  }
}

The @latest specifier is optional and will pull the latest version when new ones are available. If you don't have uvx installed you can use pipx run instead, or clone this repository locally and use the instructions in Development below.

Docker with Zotero Web API

If you want to run this MCP server in a Docker container, you can use the following configuration, inserting your API key and library ID:

{
  "mcpServers": {
    "zotero": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-e", "ZOTERO_API_KEY=PLACEHOLDER",
        "-e", "ZOTERO_LIBRARY_ID=PLACEHOLDER",
        "ghcr.io/kujenga/zotero-mcp:main"
      ],
    }
  }
}

To update to a newer version, run docker pull ghcr.io/kujenga/zotero-mcp:main. It is also possible to use the docker-based installation to talk to the local Zotero API, but you'll need to modify the above command to ensure that there is network connectivity to the Zotero application's local API interface.

Development

Information on making changes and contributing to the project.

1. Clone this repository
2. Install dependencies with uv by running: uv sync
3. Create a .env file in the project root with the environment variables above

Start the MCP Inspector for local development:

npx @modelcontextprotocol/inspector uv run zotero-mcp

To test the local repository against Claude Desktop, run echo $PWD/.venv/bin/zotero-mcp in your shell within this directory, then set the following within your Claude Desktop configuration

{
  "mcpServers": {
    "zotero": {
      "command": "/path/to/zotero-mcp/.venv/bin/zotero-mcp"
      "env": {
        // Whatever configuration is desired.
      }
    }
  }
}

Running Tests

To run the test suite:

uv run pytest

Releasing

1. Bump the version in pyproject.toml and server.json (two places: the server version and the package version).
2. Publish to PyPI with make publish.
3. Publish updated metadata to the official MCP registry with make publish-mcp, authenticating first with mcp-publisher login github if needed.

Docker Development

Build the container image with this command:

docker build . -t zotero-mcp:local

To test the container with the MCP inspector, run the following command:

npx @modelcontextprotocol/inspector \
    -e ZOTERO_API_KEY=$ZOTERO_API_KEY \
    -e ZOTERO_LIBRARY_ID=$ZOTERO_LIBRARY_ID \
    docker run --rm -i \
        --env ZOTERO_API_KEY \
        --env ZOTERO_LIBRARY_ID \
        zotero-mcp:local

Relevant Documentation

• https://modelcontextprotocol.io/tutorials/building-mcp-with-llms
• https://github.com/modelcontextprotocol/python-sdk
• https://pyzotero.readthedocs.io/en/latest/
• https://www.zotero.org/support/dev/web_api/v3/start
• https://modelcontextprotocol.io/llms-full.txt can be utilized by LLMs

Appendix: notes on Zotero's API

Item metadata is rendered from whatever fields the API returns rather than from a fixed list. Fields it omits are skipped, and fields this server doesn't know about appear under an "Other Fields" heading, so a Zotero upgrade never silently drops metadata.

The two APIs differ in one way worth knowing: the local API's quick search indexes citation keys, while the Web API does not, so searching for one returns nothing over the Web API even though the field is present in the data.