Local Zotero MCP

An MCP server that gives an LLM read-only access to your local Zotero library via the desktop app's Local API. It talks to Zotero over http://127.0.0.1:23119, so nothing leaves your machine and no API keys are required.

Requirements

• Zotero desktop app, running, with Settings > Advanced > "Allow other applications on this computer to communicate with Zotero" enabled.
• uv installed.

Quickstart

No cloning or installing required - uvx fetches the published package from PyPI and runs it. Add this to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "zotero": {
      "command": "uvx",
      "args": ["--from", "local-zotero-mcp", "zotero-mcp"]
    }
  }
}

Restart Claude Desktop and the Zotero tools should be available. uvx caches the environment after the first run, so subsequent launches are fast; add --refresh to args if you want to force-pull the latest version.

Tools

Every library-scoped tool takes an optional library argument - "user" (default) for your personal library, or a group ID to read a group library synced locally. Use list_groups to find group IDs.

The two tools below have no equivalent in the Zotero Web API - they read the attachment file resolved by get_attachment_file_path and process it locally with PyMuPDF. Both accept remove_headers_footers: when true, lines that repeat near the top/bottom of most pages (running titles, page numbers, etc.) are detected and stripped from the returned/saved text. For convert_document this only applies to output_format="txt" - PDF output preserves the original pages as-is.

Claude Plugin

This repo also ships as a Claude Code plugin bundling four Agent Skills for Zotero-based research, on top of the same MCP tools listed above:

The plugin's .mcp.json runs the server the same way as the Quickstart above (uvx --from local-zotero-mcp zotero-mcp), so it always tracks the published PyPI package.

To try it locally, clone the repo and run:

claude --plugin-dir /path/to/Local-Zotero-MCP

Then invoke a skill directly, e.g. /local-zotero-mcp:searching-zotero-library, or just ask Claude a research question about your library and let it pick the right skill.

Local development

Clone the repo and install it locally:

uv sync

Run it directly from the checkout:

uv run zotero-mcp

Point Claude Desktop at your local checkout instead of GitHub with:

{
  "mcpServers": {
    "zotero": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/Local-Zotero-MCP", "zotero-mcp"]
    }
  }
}

Run the tests:

uv run pytest

Tests mock the Zotero HTTP responses, so a running Zotero instance isn't required. To smoke-test against your own Zotero library:

uv run mcp dev zotero_mcp/server.py