A powerful Zotero AI and MCP plugin with ChatGPT, Gemini 3.6, Claude Fable 5, Claude Sonnet 5, DeepSeek V4, Grok, OpenRouter, Kimi k3, GLM 5.2, SiliconFlow, GPT
A powerful Zotero AI and MCP plugin with ChatGPT, Gemini 3.6, Claude Fable 5, Claude Sonnet 5, DeepSeek V4, Grok, OpenRouter, Kimi k3, GLM 5.2, SiliconFlow, GPT
PapersGPT is a professional-grade Zotero extension engineered in Native C++ to transform thousands of isolated PDFs into a single, high-precision knowledge engine. While standard tools struggle with single-file chat or slow, "fuzzy" embeddings, PapersGPT delivers instant, cross-document synthesis and global library search with absolute data sovereignty.
Stop waiting for embedding progress bars. Most AI tools use Python/JS scripts that lag or crash as your library grows. PapersGPT is built to handle professional-scale research. It turns massive document silos into instant knowledge bases. Whether you have 100 or 10,000+ PDFs, the experience remains lightning-fast. PapersGPT’s Enterprise-Grade Performance:
View Detailed Scalability Report & System Specs
PapersGPT moves beyond the "fuzzy" matching of standard semantic search by embeddings. Our engine understands document structure, ensuring that cross-paper comparisons are logically sound and factually accurate.
Deep Cross-Document Comparison
“Compare the sample sizes and p-values of all 50 clinical trials in my 'Cardiology' folder.”
AI synthesizes data from multiple files into structured tables or reports without losing the nuances of individual studies.
Global Library Synthesis
Ask questions across your entire Zotero library. Identify trends, evolution of terms, or conflicting evidence across 10,000+ documents simultaneously.
Verifiable Evidence (Click-to-Jump)
Accuracy is useless without proof. Every synthesis includes interactive citations. Every AI response includes precise citations; clicking on them takes you directly to the specific paper being cited.
Ensure the absolute 100% privacy and security of your local sensitive Zotero data
Support 100% offline local LLMs
PapersGPT currently supports three different ways to connect to local large language models (LLMs):
Integrated SOTA open-source LLMs – such as Gemma 4 12B and Qwen3.5 4B. These models can be downloaded and run with a single click, eliminating the need for complex installation procedures; simply selecting a model from the "Local LLM" option triggers an automatic download from Hugging Face and launches the model.
Custom OpenAI-compatible endpoint – connect to a flexible OpenAI-compatible privately deployed LLM endpoint by configuring a customized LLM API within PapersGPT.
Ollama integration – supports connections to local Ollama instances; any LLM deployed and running locally via Ollama can be connected and used simply by configuring it within PapersGPT.
Parsing, Indexing and Searching documents in Zotero are performed locally, there is no reliance on the clouds, and these operations can be run offline.
Local LLMs and local document processing ensure "zero-byte" data leakage from your device.
Chat with a Single PDF
Command + EnterCtrl + EnterChat with Multiple PDFs
Ctrl while clicking files on Windows. Hold Command while clicking files on Mac.Chat with the entire Zotero Library
For more detailed information please see here.
Support MCP which makes you can access your Zotero papers in Claude, ChatGPT desktop app, please see here in detail.
Professional SKILL provides various agents for connecting to your local Zotero library.
If you use OpenClaw, you can see the papersgpt-for-zotero skill here. You are welcome to submit any issues in this github repo or clawhub.
If you use the App of OpenAI, Claude, Cursor, Claude Code, CodeX or Gemini CLI, you can directly copy and load the SKILL.
PapersGPT is designed for professionals in Research, Legal and Finance who handle high-stakes, sensitive data:
Unlock Professional Capabilities at papersgpt.com →
Don't just search your library. Understand it. PapersGPT provides the industrial-scale intelligence your research library demands.
No open issues yet, or sync has not completed.