ZotGrep - Enhanced Zotero Library and Full-Text PDF Search

ZotGrep is a Python package that enables users to search their local Zotero library using the API and then search for full-text content within PDFs. It includes multiple output formats (CSV and Markdown) and Zotero URL integration for direct access to search results.

The general workflow involves a) including search terms for references in the Zotero library, b) full-text search among the results for a new set of keywords. The output will contain all hits among the references.

Credits: ZotGrep builds on the excellent pyzotero by Stephan Hugel.

Disclaimer: The project includes substantial parts that were vibe-coded using Claude and ChatGPT.

Features

Core Functionality

• Search Zotero library metadata (titles, authors, etc.)
• Full-text search within PDF attachments (both linked and imported files)
• Easy-to-use web interface and advanced command line interface for power users and AI agents.
• Support for both local linked files and Zotero-stored PDFs
• Context-aware text extraction with highlighted search terms
• Multiple Output Formats: Save search results to CSV or Markdown files
• CSV Export: Structured data format with comprehensive metadata for analysis
• Markdown Export: Research-friendly format with YAML frontmatter for note-taking apps
• Zotero URLs: Direct links to open items and specific PDF pages in Zotero
• Enhanced Metadata: Author names, publication years, and timestamps
• Command-line Options: Automation-friendly with argument parsing
• Interactive Output Choice: Choose between CSV, Markdown, or no file output

Installation

A slide-deck tutorial for installation and usage is also available at: franciscowilhelm.github.io/zotgrep/.

Install from PyPI

For most users, install ZotGrep from PyPI and run it as a CLI tool:

uv tool install zotgrep

Then launch it with:

zotgrep --help

If you want to use ZotGrep as a dependency inside another uv-managed project instead of as a global tool:

uv add zotgrep

Install from a source checkout

If you are developing ZotGrep or want to run the local repository in editable mode:

git clone https://github.com/franciscowilhelm/zotgrep.git
cd zotgrep
uv venv .venv
source .venv/bin/activate
uv pip install -e .

Usage

Web Interface

Launch the local web interface with:

zotgrep --web

By default the web UI runs on http://127.0.0.1:23120. You can override that with --port, for example:

zotgrep --web --port 23121

ZotGrep supports a user config file for persistent defaults. You can manage the most important settings via the web UI under General Settings. Use that page to save defaults such as linked-file paths, result limits, and context-window size. The main search page then uses those saved defaults and keeps only per-search inputs in the form. See Advanced for configuring settings via a JSON file.

Interactive Search via Command Line Interface

You can use ZotGrep via the zotgrep shell command or the module interface, which provides an interactive shell:

zotgrep

Equivalent module form:

python -m zotgrep

This will prompt you for:

• Metadata search terms (searches titles, authors, etc.)
• Full-text search terms (comma-separated list)

After displaying results, you'll be offered output format choices:

Output options:
1. CSV file (spreadsheet format)
2. Markdown file (research notes format)
3. No file output
Choose output format (1/2/3):

Direct Search via Command Line

You can specify search terms directly via command line arguments for non-interactive use:

zotgrep --zotero "career engagement"
zotgrep --zotero "career engagement" --fulltext "barriers"
zotgrep --zotero "career engagement" --metadata-only
zotgrep --zotero "career engagement" --no-abstract
zotgrep --zotero "AI ethics" --item-type "journalArticle, bookSection" --tags "privacy, fairness" --tag-match any
zotgrep --zotero "measurement invariance" --collection "Focused Review"
zotgrep --zotero '"career engagement", "career orientation"' --metadata-only

• --zotero specifies the metadata search string (e.g., title, author, etc.). It supports boolean syntax: comma or OR for alternatives, AND for conjunction (AND binds tighter than OR), and double-quoted phrases for exact phrase matching. For example, zotgrep --zotero '"career engagement", "career orientation"' --metadata-only runs one Zotero search per alternative and merges the results.
  • Phrases are verified against title, creator names, and date, but only in the default Title-Author-Year search mode; in Everything mode phrases fall back to "all words present" matching.
  • * wildcards and parentheses are not supported in metadata queries (wildcards work only in --fulltext queries); * is passed to Zotero unchanged and matches literally.
  • Plain queries without any operators behave exactly as before: they are passed verbatim to Zotero quick search, where space-separated words are implicitly ANDed as word matches (not phrases).
  • Unquoted words within boolean alternatives are also implicitly ANDed: career engagement OR motivation does not require the phrase "career engagement". The Max Results limit applies to each alternative, so the merged result count can exceed it.
• --fulltext optionally specifies the full-text search terms (comma-separated).
• --metadata-only / --no-fulltext runs only the metadata search and skips PDF/full-text processing.
• abstracts are included by default; --no-abstract omits them.
• --publication / --publication-title filters results by publication title (comma-separated for multiple). This filter is applied client-side after the Zotero API metadata query.
• --item-type / --itemtype filters by Zotero item type (comma-separated for multiple).
• --collection filters by a Zotero collection key or exact collection name.
• --tag / --tags filters by Zotero tag (comma-separated for multiple).
• --tag-match {all,any} controls whether all supplied tags are required or any single tag is enough.

This allows for scripting and automation without interactive prompts. All other output and configuration options remain available.

Example with output:

zotgrep --zotero "AI ethics" --fulltext "privacy, fairness" --csv results.csv

Example with publication filter (list via comma-separated values):

zotgrep --zotero "AI ethics" --fulltext "privacy, fairness" --publication "Nature, Science"

Example with metadata filters:

zotgrep --zotero "AI ethics" --fulltext "privacy, fairness" --item-type "journalArticle" --collection "Focused Review" --tags "privacy, fairness" --tag-match all

Output Format Options

CSV Export

Save results to CSV format for data analysis and spreadsheet applications:

# Save to specific CSV file
zotgrep --csv results.csv

# Save to CSV only (no console output)
zotgrep --csv results.csv --csv-only

Markdown Export

Save results to Markdown format for research notes and documentation:

# Save to specific Markdown file
zotgrep --md results.md
zotgrep --markdown results.md

# Save to Markdown only (no console output)
zotgrep --md results.md --md-only
zotgrep --markdown results.md --markdown-only

JSON Export

Structured JSON output is saved by default unless --no-json is used. You can also specify a filename explicitly:

zotgrep --zotero "career engagement" --json results.json
zotgrep --zotero "career engagement" --no-json

JSON and Markdown frontmatter both record the applied metadata filters inside search_details.metadata_filters.

Interactive Output Choice

When no output format is specified, the script offers an interactive menu to choose between CSV, Markdown, or no file output.

Output Formats

CSV Output Format

The CSV file includes the following columns for structured data analysis:

Notes:

• Metadata-only runs include only reference-level columns.
• Full-text columns such as pdf_filename, page_number, search_term_found, context, and zotero_pdf_url are included only when full-text hits exist.
• The abstract column is omitted when --no-abstract is used.
• CSV uses plain context; the Markdown-only highlighted variant is not written to CSV.

Markdown Output Format

The Markdown output is designed for research note-taking and literature review workflows, and is structured as follows:

• YAML Frontmatter Block: At the top, a compact YAML block stores the format version, search_details, and a summary.
• Search Summary and Reference List: A summary of the search and a numbered reference list for all papers.
• Abstracts Section: By default, a separate abstracts section appears after the reference list unless --no-abstract is supplied.
• Detailed Findings Section:
  • In full-text mode, each paper includes metadata, a per-term occurrence summary, and numbered annotation excerpts with Zotero PDF links.
  • In metadata-only mode, the file still includes the reference list and abstracts, but the detailed findings section states that no annotation-level findings were generated.
• Detailed Paper Sections in Full-Text Mode: For each paper:
  • The paper title as a heading.
  • Metadata bullets: authors, year, publication, DOI, citekey, and a direct Zotero link.
  • A Term Summary subheading with occurrence counts per search term.
  • An Annotations subheading with numbered occurrences and page-specific Zotero links.
  • A horizontal rule (---) separates each paper section.

This format is compatible with note-taking applications like Obsidian and supports direct navigation to Zotero items and PDF pages.

Example Markdown Structure:

---
zotgrep-results/v1:
  search_details:
    zotero_query: bifactor
    full_text_query:
    - psycho
    search_mode: fulltext
    search_timestamp: '2025-06-08 17:15:00'
    context_window: 2
  summary:
    total_papers_found: 3
    total_annotations_found: 17
---

# ZotGrep Results

## Search Summary

- **Search Date:** 2025-06-08 17:15:00
- **Zotero Library Query:** `bifactor`
- **Full-Text Query:** `psycho`
- **Results:** Found **17** annotations across **3** papers.

### Reference List

1. Neufeld, S., St Clair, M., Brodbeck, J., Wilkinson, P., Goodyer, I., & Jones, P. (2024). *Measurement Invariance in Longitudinal Bifactor Models: Review and Application Based on the p Factor*.
2. Bornovalova, M., Choate, A., Fatimah, H., Petersen, K., & Wiernik, B. (2020). *Appropriate Use of Bifactor Analysis in Psychopathology Research: Appreciating Benefits and Limitations*.
3. Watts, A., Poore, H., & Waldman, I. (2019). *Riskier Tests of the Validity of the Bifactor Model of Psychopathology*.

---

## Detailed Findings

### Measurement Invariance in Longitudinal Bifactor Models: Review and Application Based on the p Factor

- **Authors**: Neufeld, Sharon A. S.; St Clair, Michelle; Brodbeck, Jeannette; Wilkinson, Paul O.; Goodyer, Ian M.; Jones, Peter B.
- **Year**: 2024
- **Publication**: Psychological Assessment
- **DOI**: https://doi.org/10.1037/pas0000564
- **Citekey**: `73ZD2D7S`
- **Zotero Link**: [Open Item in Zotero](zotero://select/library/items/73ZD2D7S)

#### Term Summary

- `psycho`: 2 occurrences

#### Annotations

##### Occurrence #1, Page 8

> Thus far we have reviewed the importance of establishing longitudinal MI in bifactor models, provided guidance on MI cut-offs to employ when ordered-categorical indicators are utilized, and outlined estimator choices and missing data considerations. ...
> - Highlight on [Page 8](zotero://open-pdf/library/items/KGDL8AWR?page=8)

##### Occurrence #2, Page 18

> Psychological Assessment, 30(9), 1174–1185. https://doi.org/10.1037/pas0000564 ...
> - Highlight on [Page 18](zotero://open-pdf/library/items/KGDL8AWR?page=18)

---

# ... Additional paper sections follow the same structure ...

Quick Start Examples

Example 1: Basic Search with Interactive Output Choice

zotgrep
# Enter search terms when prompted
# Choose output format from the interactive menu

Example 2: Direct Search via Command Line (Non-Interactive)

zotgrep --zotero "deep learning" --fulltext "convolution, neural network"
# Runs search directly with specified terms, no prompts

Example 3: Direct CSV Export

zotgrep --zotero "AI ethics" --fulltext "privacy, fairness" --csv my_research_results.csv
# Results saved to CSV for data analysis

Example 4: Direct Markdown Export for Note-Taking

zotgrep --zotero "literature review" --fulltext "systematic, meta-analysis" --md literature_review.md
# Results saved to Markdown for research notes

Example 5: Silent Export (No Console Output)

zotgrep --zotero "machine learning" --fulltext "algorithm, bias" --markdown research_notes.md --markdown-only
# Only creates the Markdown file, no console output

Open Item in Zotero

zotero://select/library/items/ITEM_KEY

Open PDF at Specific Page

zotero://open-pdf/library/items/PDF_KEY?page=PAGE_NUMBER

These URLs can be clicked in spreadsheet applications or used programmatically to jump directly to relevant content in your Zotero library.

Example Output

Console Output

Reference: Machine Learning in Healthcare (Key: SMITH2023)
  Authors: Smith, John; Doe, Jane
  Year: 2023
  PDF: smith_2023_ml_healthcare.pdf
  Found 'algorithm' on Page: 15
  Context: ...The machine learning ***algorithm*** demonstrated significant improvements...
  Zotero PDF URL: zotero://open-pdf/library/items/PDF123?page=15

CSV Output

The same information is saved in structured CSV format for further analysis, reporting, or integration with other tools.

Markdown Output

Results are organized by paper with YAML frontmatter and annotations sections, perfect for research note-taking and literature review workflows.

Command Line Arguments

Output Options

• --json FILENAME: Save results to specified JSON file
• --no-json: Disable the default JSON export
• --csv FILENAME: Save results to specified CSV file
• --csv-only: Only save to CSV, suppress console output
• --md FILENAME or --markdown FILENAME: Save results to specified Markdown file
• --md-only or --markdown-only: Only save to Markdown, suppress console output

Search Term Options

• --zotero "SEARCH TERMS": Specify Zotero metadata search terms directly (e.g., "machine learning health"). Supports boolean syntax: comma or OR for alternatives, AND for conjunction (AND binds tighter than OR), and double-quoted phrases for exact matching, e.g. '"career engagement", "career orientation"'. * wildcards and parentheses are not supported here (use --fulltext for wildcards). Plain queries without operators are passed verbatim to Zotero quick search.
• --fulltext "TERM1, TERM2": Optionally specify full-text search terms as a comma-separated list (e.g., "algorithm, bias")
• --metadata-only or --no-fulltext: Skip PDF/full-text processing and return metadata-only results
• --no-abstract: Omit abstracts from output
• --publication "TITLE1, TITLE2" or --publication-title "TITLE1, TITLE2": Filter results by publication title (comma-separated list). Example: "Nature, Science". This filter is currently applied client-side after the Zotero API metadata query, because the Zotero Web API does not expose a publication-title search parameter.
• --item-type "TYPE1, TYPE2" or --itemtype "TYPE1, TYPE2": Filter by Zotero item type. Example: "journalArticle, book"
• --collection "COLLECTION": Filter by Zotero collection key or exact collection name. Example: "ABCD1234" or "Focused Review"
• --tag "TAG1, TAG2" or --tags "TAG1, TAG2": Filter by Zotero tags. Example: "privacy, fairness"
• --tag-match {all,any}: Control how multiple tags are matched. all requires every supplied tag; any accepts at least one.

Other Options

• --config CONFIG: Path to configuration file (JSON format)
• --base-path PATH: Override base attachment path
• --max-results N: Maximum results per metadata search alternative (default: config file, ZOTERO_MAX_RESULTS, or 100)
• --context-window N: Context sentence window size (default: config file, ZOTERO_CONTEXT_WINDOW, or 2). A window of 2 means 2 sentences before and after the keyword is found will be returned. Larger window sizes will return more sentences. Sentence splitting uses the Zotero item language when available, with a built-in fallback if no language-aware tokenizer is available at runtime.
• --port PORT: Port for the local web interface when using --web (default: 23120)
• --version: Show version information
• --help: Show help message

Environment Variables

• ZOTGREP_CONFIG_PATH: Use a custom user config file path
• ZOTERO_BASE_ATTACHMENT_PATH: Base directory for linked-file attachments
• ZOTERO_PUBLICATION_TITLE_FILTER: Filter results by publication title (comma-separated list). Example: Nature, Science
• ZOTERO_ITEM_TYPE_FILTER: Filter by Zotero item type (comma-separated list). Example: journalArticle, book
• ZOTERO_COLLECTION_FILTER: Filter by Zotero collection key or exact collection name
• ZOTERO_TAG_FILTER: Filter by Zotero tags (comma-separated list). Example: privacy, fairness
• ZOTERO_TAG_MATCH_MODE: Control multi-tag matching with all or any

Use Cases

Research and Literature Review

• CSV Export: Create structured datasets for systematic literature reviews and meta-analyses
• Markdown Export: Generate research notes with proper citations and page references
• Direct Zotero Integration: Jump directly to source materials from search results

Academic Writing

• Evidence Collection: Quickly locate and cite relevant passages with page-specific references
• Collaborative Research: Share search results in both structured (CSV, JSON) and readable (Markdown) formats
• Literature Synthesis: Build comprehensive literature reviews with organized annotations

Knowledge Management

• Research Databases: Create searchable databases of research findings in CSV format
• Note-Taking Integration: Import Markdown results into Obsidian, Notion, or other note-taking apps
• Concept Tracking: Monitor mentions of specific concepts across your entire literature collection
• Citation Networks: Build interconnected knowledge bases with contextual references

Workflow Integration

• Data Analysis: Use CSV exports with R, Python, or Excel for quantitative literature analysis
• Documentation: Generate Markdown reports for research documentation and sharing
• Reference Management: Seamlessly integrate with existing Zotero workflows
• Agentic Workflows: Let AI agents use the CLI and use JSON (highly structured and machine-readable, less human-readable) output formats.

Troubleshooting

Common Issues

1. PDF not found errors: For linked files, verify BASE_ATTACHMENT_PATH is correctly set; for Zotero-stored files, verify the attachment is available through Zotero
2. API connection failures: Check Zotero credentials and internet connection
3. Empty search results: Try broader search terms or check PDF text extraction
4. File encoding issues: Both CSV and Markdown files use UTF-8 encoding for international characters
5. Markdown formatting issues: Special characters are automatically escaped in Markdown output
6. Output format conflicts: Cannot use both --csv-only and --md-only simultaneously

Debug Tips

• Check Zotero sync status
• For linked files, verify PDF files are accessible at the specified paths
• Test with simple search terms first
• Review console output for detailed error messages
• For Markdown issues, check that special characters are properly handled
• Use --help to see all available command-line options

Output Format Selection Guide

Choose CSV when:

• Performing quantitative analysis of search results
• Importing data into spreadsheet applications
• Conducting systematic literature reviews requiring structured data
• Integrating with data analysis tools (R, Python, etc.)

Choose Markdown when:

• Taking research notes and building knowledge bases
• Using note-taking applications like Obsidian or Notion
• Creating readable research documentation
• Building interconnected literature reviews
• Sharing results in a human-readable format

Advanced

Modifying settings via JSON file

Additional to configuring basic settings via the Web UI, settings can be modified in these ways:

• manually by creating that JSON file
• by pointing to another file with --config PATH or ZOTGREP_CONFIG_PATH

The recommended file path is:

~/.config/zotgrep/config.json

Typical persistent settings include:

{
  "zotero_user_id": "0",
  "zotero_api_key": "local",
  "library_type": "user",
  "base_attachment_path": "/path/to/your/linked/pdfs",
  "max_results_stage1": 100,
  "context_sentence_window": 2,
  "item_type_filter": ["journalArticle"],
  "collection_filter": "Focused Review",
  "tag_filter": ["privacy", "fairness"],
  "tag_match_mode": "all"
}

If you use only Zotero-stored files, leave base_attachment_path empty.

Environment variables override config-file values at runtime. The most relevant ones are:

export ZOTERO_BASE_ATTACHMENT_PATH='/path/to/your/zotero/attachments'
export ZOTGREP_CONFIG_PATH='/path/to/custom/config.json'
export ZOTERO_ITEM_TYPE_FILTER='journalArticle,book'
export ZOTERO_COLLECTION_FILTER='Focused Review'
export ZOTERO_TAG_FILTER='privacy,fairness'
export ZOTERO_TAG_MATCH_MODE='any'

Full-text backend comparison: pypdfium2 vs. Zotero index (xpdf)

ZotGrep offers two Stage 2 full-text backends, selectable with --fulltext-source (only in the CLI, the web interface always uses pdf):

• pdf (default) — ZotGrep downloads each PDF attachment and extracts text with pypdfium2. Provides page-level results and context windows.
• zotero-index (experimental) — ZotGrep queries the full-text index that Zotero itself maintains (built internally using xpdf). No download is required, but page-level information is unavailable and not all attachments may be indexed.

The table below shows results across five searches against the same Zotero library (--zotero proactiv, metadata search mode: titleCreatorYear). Hits = total annotation matches returned (context windows, not unique papers).

Takeaways:

• pypdfium2 found 12–37% more hits per query in four out of five cases, likely because Zotero's index does not cover all pages or all attachments or misses hits due to poorer PDF processing.
• pypdfium2 was faster in every query, and significantly so for high-hit queries (daily: 2× faster). The overhead of downloading and parsing PDFs is more than offset by avoiding sequential index API calls.

Use zotero-index only if you cannot access the PDF files directly. For all other cases the default pdf backend gives better coverage and lower latency.

Testing

Run the test suite to verify functionality:

uv run --group test python -m pytest

This will test:

• Zotero URL generation
• CSV export functionality
• Sample data processing

License

This project is open source published, like Zotero itself, under a GPL license. Please refer to the license file for details.

Acknowledgments

ZotGrep depends on several upstream open-source projects. In particular:

• pyzotero for Zotero API access
• Flask for the local web interface
• pypdfium2 and PDFium for PDF text extraction
• pySBD for sentence boundary detection
• PyYAML for YAML serialization in Markdown exports

See NOTICE for license attributions and upstream license links.

Contributing

Contributions are welcome! Please feel free to submit issues, feature requests, or pull requests.