zotero-notebooklm-skill: The Missing Link in the AI Research Stack

How a lean Claude Code skill automates the leap from curated citations to synthesized insights.

6 min read · orhoncan/zotero-notebooklm-skill

A wide-angle illustration of a massive, complex loom weaving raw threads representing PDFs into a structured tapestry representing an AI notebook. — The skill acts as an orchestrator, weaving structured local data into cloud-based synthesis engines.

Key Takeaways

This tool uses the Model Context Protocol to bridge local Zotero databases with cloud-based NotebookLM instances.
The skill preserves intellectual context by uploading researcher annotations as distinct data sources from the original PDFs.
A bidirectional loop allows the AI to identify literature gaps and save new research papers directly back to the local Zotero library.
The architecture replaces manual document uploads with an automated orchestration layer managed through Claude Code.

The End of the Manual Upload

Researchers face a constant friction point in the modern AI stack. Papers and metadata live in structured, local databases like Zotero, while the actual synthesis and reasoning happen in isolated AI environments like Google NotebookLM.

Bridging this gap usually requires tedious manual labor. Users download PDFs, drag them into web interfaces, and copy-paste their notes. In this manual transition, the structured metadata is often lost, and the context becomes flattened.

MCP as the Universal Joint

The zotero-notebooklm-skill repository introduces a different approach. Rather than building a monolithic application, it acts as a lightweight orchestration layer for Claude Code.

An interactive architecture diagram showing Claude Code in the center acting as a relay. On the left is a node labeled 'Zotero MCP (Local SQLite)'. On the right is a node labeled 'NotebookLM MCP (Cloud)'. An animated data packet flows from the Zotero node

By leveraging the Model Context Protocol (MCP), the skill commands two independent servers that are entirely unaware of each other. It fetches data from a local Zotero SQLite database and pipes it directly into a cloud-based NotebookLM instance.

1. zotero.get_collections
2. notebooklm.notebook_create
3. zotero.get_collection_items
4. notebooklm.source_add

Beyond Files: Syncing the Researcher's Ghost

Moving files is trivial, but moving knowledge requires nuance. This skill explicitly separates the raw text of a paper from the researcher's own highlights and annotations.

A close-up illustration of a physical book where printed text is faint, while handwritten margin notes are bold and being lifted off the page by a mechanical tool. — Separating the author's text from the researcher's annotations provides the AI with critical context.

By uploading these annotations as distinct context sources, the AI can differentiate between the original author's claims and the reader's ongoing hypotheses. The system syncs not just the document, but the researcher's intellectual footprint.

Closing the Loop with Deep Research

The pipeline is not a one-way street. The skill includes an expansion command that prompts the AI to identify gaps in the current literature collection.

Traditional Workflow	MCP Orchestration Pipeline
Manual PDF uploads to a web UI	Automated syncing via CLI commands
Annotations merged with source text	Annotations uploaded as distinct context sources
One-way data flow to the AI	Bidirectional sync writing new papers back to local Zotero
Limited by browser memory and UI	Headless execution leveraging full MCP capabilities

Once gaps are identified, the AI conducts deep research, finds missing papers, and uses Zotero's local creation tools to save them back to the user's database. This completes a bidirectional loop, transforming the AI from a passive reader into an active research assistant.