zotero-mcp-code

The Zotero MCP Code Execution skill is a powerful utility designed for researchers and knowledge workers who manage extensive academic libraries. By shifting the search process from direct API tool calls to a Python-based code execution environment, it circumvents the limitations of standard tool interactions. This skill allows you to fetch large datasets of over 100 items at once, perform complex filtering (such as by DOI, date range, or publication type), and apply sophisticated ranking algorithms directly within the execution sandbox before presenting findings to the user. It is ideal for users who need to conduct systematic literature reviews, search for specific papers across multiple topics simultaneously, or perform multi-angle queries that would otherwise overwhelm a standard chat interface.

• Performs automated multi-strategy searches combining semantic, keyword, and tag-based discovery in one request.

• Implements internal deduplication logic to ensure results are clean and unique.

• Provides relevance ranking to prioritize the most significant documents for the user.

• Handles large result sets in code to prevent context bloat and minimize system crash risks.

• Supports custom Python filtering for granular control over library subsets, such as filtering for recent journal articles with specific metadata.

• Integrates seamlessly with Claude Code to provide an augmented research assistant experience.

• Input: User natural language research queries or specific criteria like author, tag, or publication date.

• Output: A highly relevant, filtered, and formatted list of Zotero library items ready for immediate review.

• Intended for users familiar with bibliographic management who require programmatic access to their Zotero data.

• Optimized for deep research, literature synthesis, and complex knowledge retrieval tasks.

• Operates best when the user has existing libraries in Zotero managed through the standard MCP interface.

• Ensures token efficiency by processing data locally in code and returning only the top N results to the LLM context.