Let your AI assistant rummage through your Zotero library locally

What it does

The plugin runs an integrated MCP server inside Zotero 7, letting AI clients such as Claude Desktop, Cherry Studio, or Cursor query your local reference library over Streamable HTTP. It exposes roughly twenty tools for full-text search, metadata retrieval, annotation analysis, and even write-back operations like tagging or note creation, all while keeping data on your machine.

The interesting bit

Instead of the usual fragile dance between a standalone server and a thin client plugin, the entire MCP server lives inside the Zotero plugin itself. That unified architecture removes a whole layer of process management and keeps your PDFs and notes strictly local by design.

Key highlights

• Twenty MCP tools span five categories: advanced search, collection browsing, semantic lookup, a cached full-text database, and write operations.
• Semantic search uses embedding vectors via OpenAI or Ollama, stored in SQLite-vec, to surface conceptually related papers even across languages.
• Content extraction offers four modes—minimal, preview, standard, and complete—so the AI can request anything from a citation snippet to an entire PDF transcript.
• Write operations (notes, tags, metadata updates) can be toggled off in preferences if you prefer your assistant to be read-only.
• Auto-configuration generates the necessary JSON for supported clients, so you do not have to hand-edit connection details.

Verdict

This is built for researchers and students who already curate their lives in Zotero and want an AI assistant that actually knows their reading history. If you do not use Zotero, or you are content with uploading PDFs to cloud chatbots, this bridge will not change your life.