Your Zotero library finally talks back to Claude

What it does

Zotero MCP exposes your entire research library—papers, annotations, collections, tags—to any Model Context Protocol client. Ask Claude to summarize a paper, find everything you’ve tagged “#Arm”, or extract PDF highlights with page numbers. It works locally (no API key) or via Zotero’s web API, and includes a standalone CLI for when you just want to grep your bibliography from the terminal.

The interesting bit

The semantic search is where it gets weirdly useful. It embeds your library with a local MiniLM model (or OpenAI/Gemini if you’re fancy), then lets you query by concept—“papers like this abstract” rather than keyword matching. The optional Scite integration also surfaces retraction alerts and citation tallies (supporting, contrasting, mentioning) without needing a Scite account.

Key highlights

• Runs fully offline in local mode; hybrid mode reads local, writes via web API
• Optional extras keep the base install lightweight—no ML bloat unless you opt into semantic or pdf
• Can add papers by DOI with auto-fetched metadata and open-access PDF cascade (Unpaywall → arXiv → Semantic Scholar → PMC)
• Includes zotero-cli with short aliases (s, g, ann, coll) for terminal-native workflows
• Auto-configures Claude Desktop; also supports ChatGPT, Cherry Studio, Cursor, Chorus

Caveats

• Semantic search requires manual database builds (update-db) or scheduled syncs; not real-time
• Full-text extraction and PDF outline parsing need separate optional installs
• Zotero 7+ required for local API with full-text access

Verdict

Researchers drowning in PDFs they can’t remember acquiring should try this. If your reference manager is already a well-organized shrine you visit daily, it’s probably overkill.