Upgrade StorageZotPilot: MCP server for semantic search, classification, and literature review drafting from your Z
Hi,
Sharing a tool I built: ZotPilot, an MCP server that connects your local Zotero library to AI agents (Claude Code, Codex CLI, opencode, etc.).
My use case: 300+ papers, Zotero's built-in search wasn't cutting it anymore. I'd know I'd read about something but couldn't find the paper because I used different words than the authors. My collections got increasingly messy since papers often fit multiple categories.
ZotPilot reads your local `zotero.sqlite` and PDF storage, chunks the text, and builds a vector index. Your AI agent can then search semantically (passage-level, not just title matching), reorganize your library based on actual paper content, draft literature review paragraphs with real citations, and manage tags and collections.
Embedding options: Gemini API, DashScope, or fully local/offline. 32 MCP tools. Python 3.10+, installed via `uv`. macOS, Linux, Windows.
PDFs are processed locally. Cloud embedding mode sends only text chunks to the API. Fully offline mode available.
One CLI command to register. No Zotero plugin required.
GitHub: https://github.com/xunhe730/ZotPilot
Bug reports and feedback welcome.
Sharing a tool I built: ZotPilot, an MCP server that connects your local Zotero library to AI agents (Claude Code, Codex CLI, opencode, etc.).
My use case: 300+ papers, Zotero's built-in search wasn't cutting it anymore. I'd know I'd read about something but couldn't find the paper because I used different words than the authors. My collections got increasingly messy since papers often fit multiple categories.
ZotPilot reads your local `zotero.sqlite` and PDF storage, chunks the text, and builds a vector index. Your AI agent can then search semantically (passage-level, not just title matching), reorganize your library based on actual paper content, draft literature review paragraphs with real citations, and manage tags and collections.
Embedding options: Gemini API, DashScope, or fully local/offline. 32 MCP tools. Python 3.10+, installed via `uv`. macOS, Linux, Windows.
PDFs are processed locally. Cloud embedding mode sends only text chunks to the API. Fully offline mode available.
One CLI command to register. No Zotero plugin required.
GitHub: https://github.com/xunhe730/ZotPilot
Bug reports and feedback welcome.