Skip to content

Using with Claude Code

Agent-native interface (0.3.0+)

zot auto-emits JSON envelopes when stdout is not a TTY, so Claude Code gets parseable output without --json. Exit codes are typed (validation, auth, not-found, network, conflict), every envelope carries a meta.request_id, and mutating commands support --dry-run and --idempotency-key. Full contract: Agent Interface.

Install the Skill

Copy the zotero-cli skill so Claude Code automatically recognizes literature-related requests:

cp -r skill/zotero-cli ~/.claude/skills/

How It Works

With the skill installed, Claude Code automatically uses zot commands when you ask about papers:

Search my Zotero for single cell papers
→ Claude runs: zot --json search "single cell"

Show me details of this paper
→ Claude runs: zot --json read ABC123

Export BibTeX for these papers
→ Claude runs: zot export ABC123

Create a collection for my ICML submission
→ Claude runs: zot collection create "ICML 2026"

Collection Workflow

Scope any deep search or question to a Zotero collection. A typical research workflow with Claude Code:

  1. Create a collection for your project
  2. Add papers to it — in the Zotero app, or with zot collection move
  3. Ask questions — no index to build, results are always fresh
Create a collection called "llm-safety" for my alignment papers
→ Claude runs: zot collection create "llm-safety"

What methods do these papers use for reward hacking detection?
→ Claude runs: zot ask "reward hacking detection methods" --collection llm-safety
  and synthesizes a cited answer from the returned evidence pack

Shell Completions

Enable tab completions for faster CLI use:

zot completions zsh >> ~/.zshrc

zot completions bash >> ~/.bashrc

zot completions fish > ~/.config/fish/completions/zot.fish