54yyyu/zotero-mcp v0.2.0 on GitHub

Architecture

Split server.py (4,800 lines) into tools/ subpackage — search, retrieval, annotations, write, connectors, and shared helpers are now separate modules. server.py is a 109-line re-export shim.
Removed _ServerModule sys.modules hack — tool modules use module-level attribute access; tests patch canonical locations directly.
Optional dependency groups — [semantic] (ChromaDB, embeddings), [pdf] (PyMuPDF, EPUB), [all]. Base install is lightweight with no ML dependencies.

Deduplicated 7 item-formatting functions into single format_item_result() with configurable abstract length, tags, and extra fields.
Extracted _normalize_limit() helper replacing 12 copy-pasted blocks.
Consolidated suppress_stdout(), _strip_xml_tags(), and creator formatting duplicates.
Modernized typing in 5 modules: Optional[X] → X | None, Dict → dict, List → list.
Removed dead code across 7 modules.

Stale embedding model detection — ChromaDB collections created with a deprecated model (e.g., text-embedding-004) are now auto-detected and recreated.
Bare except: clauses — replaced with specific exception types.
PDF outline import order — defers PyMuPDF import until after attachment check.
Suppressed noisy pdfminer warnings during PDF text extraction.

README documents optional extras ([semantic], [pdf], [all]), write operations, and embedding model troubleshooting.

Full Changelog: https://github.com/54yyyu/zotero-mcp/blob/main/CHANGELOG.md