Upgrade Storage[New Plugin] Semantic Tagger: auto-tag new items from your existing tag library using Claude
Hi all,
I'd like to announce Semantic Tagger, a plugin for Zotero 7+ that automatically applies tags to newly added papers, using only tags that already exist in your library.
When an item is added (import, browser connector, Add by Identifier, or a PDF arriving via Find Full Text), the plugin reads the title, abstract, and optionally the PDF text and asks Anthropic's Claude API to pick the best-matching tags from your existing tag list. It never invents new tags: everything the model returns is validated against your library before being written, so the worst case is a wrong tag from your own vocabulary on one item. You can also batch-tag any selection via right-click → "Semantic Tagger: Tag Selected Items".
Features:
* Tags only from your existing library — non-destructive by design
* Strictness slider: from 2–4 core-topic tags up to 6–12 broader ones
* Model picker: once your API key is set, it lists the models your key can actually access
* Synonym groups (e.g. teach it that CH4 = methane = natural gas)
* PDF text via Zotero's own full-text cache, with an optional Python/PyMuPDF fallback; works alongside file-moving plugins like ZotMoov
* Session token budget with warnings, and an automatic pause if your API credits run out
You'll need your own Claude API key from console.anthropic.com (pay-per-use; a typical paper costs a fraction of a cent). The README documents exactly what data is sent and how the key is stored — please read the Privacy & Security section before using it.
Download and documentation: https://github.com/roey-angel/zotero-semantic-tagger
Direct XPI: https://github.com/roey-angel/zotero-semantic-tagger/releases/download/v0.1.0/semantic-tagger.xpi
This is a first public release (v0.1.0), developed and tested on Linux with Zotero 7; macOS and Windows should work but are untested — the Python PDF path in particular is macOS/Linux only. Feedback and bug reports are very welcome via the GitHub issue tracker.
Disclosure: The plugin was written with substantial help from Claude (Anthropic's model), which is also credited as co-author in the commit history.
I'd like to announce Semantic Tagger, a plugin for Zotero 7+ that automatically applies tags to newly added papers, using only tags that already exist in your library.
When an item is added (import, browser connector, Add by Identifier, or a PDF arriving via Find Full Text), the plugin reads the title, abstract, and optionally the PDF text and asks Anthropic's Claude API to pick the best-matching tags from your existing tag list. It never invents new tags: everything the model returns is validated against your library before being written, so the worst case is a wrong tag from your own vocabulary on one item. You can also batch-tag any selection via right-click → "Semantic Tagger: Tag Selected Items".
Features:
* Tags only from your existing library — non-destructive by design
* Strictness slider: from 2–4 core-topic tags up to 6–12 broader ones
* Model picker: once your API key is set, it lists the models your key can actually access
* Synonym groups (e.g. teach it that CH4 = methane = natural gas)
* PDF text via Zotero's own full-text cache, with an optional Python/PyMuPDF fallback; works alongside file-moving plugins like ZotMoov
* Session token budget with warnings, and an automatic pause if your API credits run out
You'll need your own Claude API key from console.anthropic.com (pay-per-use; a typical paper costs a fraction of a cent). The README documents exactly what data is sent and how the key is stored — please read the Privacy & Security section before using it.
Download and documentation: https://github.com/roey-angel/zotero-semantic-tagger
Direct XPI: https://github.com/roey-angel/zotero-semantic-tagger/releases/download/v0.1.0/semantic-tagger.xpi
This is a first public release (v0.1.0), developed and tested on Linux with Zotero 7; macOS and Windows should work but are untested — the Python PDF path in particular is macOS/Linux only. Feedback and bug reports are very welcome via the GitHub issue tracker.
Disclosure: The plugin was written with substantial help from Claude (Anthropic's model), which is also credited as co-author in the commit history.