zotero-skill

[jasonzou]

Thanks for the plugin and saved lots of time!!!!
I added a small enhancement:

• can import pdf - via 23119 api
• create an agent skill using the plugin
• my openclaw runs in a docker; and zotero runs in the host machine.

[cookjohn]

Thanks a lot for this contribution — PDF import is one of the most-requested capabilities, and the agent skill + Python CLI are a nice demonstration of driving the plugin end-to-end (including from a Docker container). A few changes are needed before we can merge:

1. Write gating (required): import_pdf modifies the library, but the new connector tools are added after the write-tool filter and marked "always available", and the call handler has no write.enabled check. All other write operations (write_note, create_collection, …) are gated both in tools/list and at call time. Please gate import_pdf the same way so users who disabled write operations aren't surprised.
2. Dead code: handlePdfUpload() in streamableMCPServer.ts is defined but never routed or called. Please remove it (or wire it into the HTTP server if you intended an upload endpoint).
3. Personal metadata: zotero-skill/_meta.json contains a marketplace ownerId/publishedAt — please drop this file from the PR.
4. list_collections duplication: the MCP server already exposes get_collections (and collection write ops since v1.4.6). Could you either reuse it in the skill or explain why the Connector-API variant is needed? Two tools returning collections with different shapes will confuse clients.
5. Placement: we'd prefer the skill live under examples/zotero-skill/ (with a README pointer) rather than at repo root — keeps the plugin source tree focused. Open to discussion.
6. Heads-up — overlap with feat: write_item 新增 import action — 支持将本地文件导入为 Zotero 附件 #76: write_item just gained an import action on main (local filePath → attachment, ships next release). Your base64 import_pdf still adds value for remote/container clients that cannot write to the host filesystem, but please coordinate the docs/skill so the two paths are clearly distinguished.

Smaller notes: callZoteroImport hardcodes port 23119 while the other two tools accept a port arg; and base64-encoding PDFs through JSON-RPC inflates payloads ~33%, so a documented size limit or a clearer error for huge files would help. (We verified fetch/atob are available in Zotero's plugin sandbox, so the platform usage is fine — though Zotero.HTTP.request is the convention elsewhere in the codebase.)

Also, could you fix the PR title typo ("zotoer" → "zotero")? Happy to merge once the gating and cleanup items are addressed.

[cookjohn]

Hi @jasonzou — following up: the June review requested six changes (gating import_pdf behind write.enabled, removing the unused handlePdfUpload, dropping zotero-skill/_meta.json, resolving the list_collections duplication, moving the skill under examples/, and the title typo). The branch merged upstream main in August but none of the review items have corresponding changes yet.

Also worth noting: the landscape has moved since June — write_item now has a local-file import action (v1.5.0) and add_by_identifier landed in v1.6.0, so the skill's docs would want updating either way.

If you'd like to continue, please respond or update within two weeks (by 2026-09-17) — happy to help with any of the items. If we don't hear back we'll close this PR to keep the queue tidy; you can reopen anytime, and the work remains referenced. The agent-skill idea itself is still welcome, especially as an examples/ contribution.