Zotero integration with LM Studio

Below are three reliable ways to wire Zotero → LM Studio. Pick A (quick, no-code), B (“live” access via MCP), or C (scripted, for full control).

A) Quick, no-code: export notes/PDFs from Zotero and use LM Studio’s “Chat with Documents”

1. In Zotero, export the papers’ notes/annotations to Markdown or TXT (plugins below) and/or export PDFs.
   • Plugins commonly used for Markdown/TXT export: mdnotes and Better Notes.
2. In LM Studio, open a chat and attach the exported files (.pdf/.docx/.txt) so the model can retrieve from them (RAG). For long files, LM Studio will auto-RAG.

Pros: no coding; works fully offline.

Cons: you re-export when your Zotero notes change; no library search from inside LM Studio.

B) “Live” integration via MCP (best balance)

LM Studio ≥ v0.3.17 can host Model Context Protocol (MCP) tools. Several community MCP servers expose your Zotero library (search, read metadata, fetch annotations) to local LLMs.

Steps (example with zotero-mcp):

1. Enable Zotero’s local API (Zotero 7): Preferences → Advanced → API → Enable local API. This is Zotero’s built-in local HTTP API (read-only) introduced for v7.
2. Install an MCP Zotero server (pick one):
   • pip install git+https://github.com/54yyyu/zotero-mcp.git → then zotero-mcp setup (docs show local and Web API modes, plus annotation tools).
   • Alternatives exist (kaliaboi/mcp-zotero, swairshah/zotero-mcp-server).
3. In LM Studio, add the MCP server under Use MCP Servers (Settings → MCP) so your model can call tools like search library, get item metadata, get notes/annotations.

Pros: query your Zotero library from the chat; fetch annotations on demand; stays local.

Cons: community projects (not official Zotero); you must keep Zotero running with the local API enabled.

C) Scripted pipeline (LM Studio OpenAI-compatible API + Zotero API)

If you prefer your own control loop (e.g., “search Zotero → stuff results into the prompt → ask the model”), combine Pyzotero with LM Studio’s OpenAI-compatible local server.

Outline:

1. Start LM Studio’s local server (default http://localhost:1234/v1).
2. Use Pyzotero to pull CSL-JSON/metadata/fulltext (local or web API).
3. Insert retrieved snippets (with item keys) into the model prompt or a RAG chunker.

References: LM Studio OpenAI-compat docs; Pyzotero quickstart; Zotero Web API v3.

What I recommend (practical setups)

• If you want “ask my library” inside LM Studio: do B (MCP). It’s the smoothest interactive experience: search collections, pull notes/annotations, then ask the model to synthesize or draft.
• If you just need the model to read a few papers with your highlights: do A (export + attach) with mdnotes/Better Notes and the PDFs.
• If you need a programmable workflow (e.g., batch literature reviews, auto-inserting citekeys): do C and optionally add Better BibTeX for stable citekeys and a local citation-picker endpoint (CAYW).

Notes & gotchas

• Local API must be on in Zotero 7 (port 23119). If a plugin blocks it, re-enable or free the port; the “connector/ping” and config entries are documented in forum/dev threads.
• CSL-JSON from the Web API doesn’t include tags; if you need tags in exports, use other formats or plugins.
• For annotation access, the Zotero MCP docs explain using local text extraction when needed; Better BibTeX can help with citekeys/annotation flows.
• LM Studio’s “Chat with Documents” currently lists .pdf/.docx/.txt; Markdown works fine if saved as .txt (or just attach the PDFs).

Key sources

• LM Studio: Chat with Documents; OpenAI-compatible API; MCP host support.
• Zotero: Web API v3; Local API in Zotero 7; Pyzotero docs.
• Zotero → Markdown/notes: mdnotes; Better Notes.
• MCP Zotero servers: 54yyyu/zotero-mcp (guide & tooling); alternatives.
• Better BibTeX (citekeys, CAYW/HTTP picker).

Start your setup today and chat with your documents. Reach out if you need assistance.