Skip to content

Zotero Knowledge MCP

Turn selected Zotero PDF attachments into a local, searchable knowledge library. Read extracted text beside the PDF, correct it when needed, and give an AI client explicit access through the Model Context Protocol (MCP).

Current release: 0.2.0-preview.1. Requires Zotero 10.0.x. Native acceptance has used Windows with Zotero 10.0.6–10.0.8; equivalent macOS and Linux acceptance is pending. This is a preview with manual installation and upgrades.

Download XPI SHA-256 checksum

Release notes and all versions

Current download: 0.2.0-preview.1 for Zotero 10.0.x. The source, release XPI and checksum are public; no GitHub account is needed to download them.

Start here

  1. Install the plugin from the release XPI.
  2. Process your first document with a supported parsing service.
  3. Configure semantic search if you want model-backed retrieval.
  4. Connect an AI client if you want external tools to search and read it.

End users do not need Node.js, npm, Python or a separate background server. Parser and model providers may require accounts, API credentials and paid quota.

What you can do

Task Guide
Organize selected papers and synchronize Zotero collections Knowledge libraries
Find papers and passages with text or semantic search Search
Read Markdown, follow citations and save text corrections Read and edit
Authorize local AI clients to use your knowledge libraries MCP connection
Understand external requests, credentials and backups Data and privacy
Diagnose configuration and processing problems Troubleshooting

Preview boundaries

Zotero remains responsible for your original library and attachments. The plugin stores derived content locally and uses configured external services for parsing, embeddings and optional reranking. Public remote MCP access, plugin cloud sync, direct import of arbitrary parser output and integrated full backup/restore are not available. A 1,000-paper full-workflow acceptance is still pending.

Plugin author: lianghx123. Source and releases are publicly accessible.