QuickCat — AI-Enhanced Metadata Agent

Modular AI Agent Toolbox for professional library cataloging. QuickCat automates the mechanical aspects of cataloging (harvesting, cleaning, validation) while augmenting the intellectual aspects (original cataloging, summary generation, authority grounding).

Built as a collection of Claude skills — each skill is a focused, self-contained unit with its own SKILL.md, scripts, references, and assets.

---

Skills

Skill	Purpose	Key Script
copy-cataloger	Harvest MARC records from Z39.50/SRU (LOC, BL, NLS, OCLC). ISBN-first search hierarchy, Green/Yellow/Red consensus engine, tie-breaker protocol.	harvest_orchestrator.py
vision-to-marc	Multimodal OCR of title page photos → MARC 21 fields via Claude Vision API.	image_to_marc.py
authority-grounder	Validate LCSH/LCNAF headings, inject $0 URI subfields from id.loc.gov, apply ISBD punctuation.	authority_lookup.py
brief-to-full-enhancer	Generate MARC 520 (Summary) and 505 (Contents Note) using Claude.	enhance_record.py
batch-cleaner	Profile-driven tag deletion (9XX, vendor fields), Unicode NFC normalization, Leader byte fix, 003 stamping.	batch_clean.py
marc-importer	Ingest ISO-2709 binary or Excel/CSV → clean → validate → import-ready .mrc.	import_pipeline.py
record-rollback	Browse the transaction journal (.quickcat.log) and restore any record to a prior state.	rollback.py
url-checker	Async HTTP HEAD validation of all 856 $u electronic access URLs.	check_856.py
marc-exporter	Export validated records to ISO-2709 with productivity metrics (original vs. copy-cataloged, AI-tagged counts).	export.py

---

Installation

git clone https://github.com/codefzer/QuickCat.git
cd QuickCat
pip install -r requirements.txt

Requirements

• Python 3.10+
• An Anthropic API key for vision and generation skills
• Z39.50/SRU network access (institutional IP or credentials for OCLC/BL/NLS)

---

Configuration

Copy the environment template and set your credentials:

# Required for vision-to-marc and brief-to-full-enhancer
export ANTHROPIC_API_KEY=sk-ant-...

# Required for Z39.50 sources that need authentication (e.g. OCLC)
export QC_Z3950_USER=your_username
export QC_Z3950_PASS=your_password

All server endpoints, timeouts, retry policy, consensus thresholds, and Excel column mappings are in config.json. Edit it for your institution before first use — in particular set org_code in skills/batch-cleaner/assets/default-profile.json to your MARC organization code.

---

Quick Start

Copy catalog a book by ISBN

python3 skills/copy-cataloger/scripts/harvest_orchestrator.py --isbn 9780743273565

Catalog from a title page photo

python3 skills/vision-to-marc/scripts/image_to_marc.py --image title_page.jpg --type book

Clean a vendor MARC file before ILS import

python3 skills/batch-cleaner/scripts/batch_clean.py vendor_records.mrc --org-code MYLIB

Import an acquisitions spreadsheet

python3 skills/marc-importer/scripts/import_pipeline.py acquisitions.xlsx --type book

Validate subject headings and inject LC URIs

python3 skills/authority-grounder/scripts/authority_lookup.py records.mrc

Undo a change

python3 skills/record-rollback/scripts/rollback.py --list --log .quickcat.log
python3 skills/record-rollback/scripts/rollback.py \
    --record-id "001:ocn12345678" --timestamp "2026-03-18T14:32:00Z" \
    --mrc records.mrc --log .quickcat.log

---

See SKILLS-OVERVIEW.md for full workflow scenarios, technical standards, shared resource reference, and configuration details.

---

License

MIT License — see LICENSE for details.