Zotero MCP acts as a powerful bridge between a personal research library and AI assistants like Claude or ChatGPT. It allows researchers to interact with their Zotero collections through natural conversation, making it easy to find specific papers, get summaries of articles, or analyze citations without manually digging through folders. By turning a static library into a dynamic knowledge base, it helps users quickly organize and retrieve their academic work using simple, everyday language. Beyond basic search, this tool can dive deep into the content of a library by extracting and searching through PDF annotations, notes, and full-text content. It supports both local access for offline privacy and a web API for cloud-based libraries, ensuring flexibility regardless of where the data is stored. For those managing large volumes of research, it offers automated sync schedules and a smart update system that preserves configurations while keeping the tool compatible with the latest AI client developments. For developers and power users, the MCP's standout feature is its AI-powered semantic search, which uses vector-based similarity to find research based on concepts rather than just keywords. It integrates seamlessly with multiple embedding models—including free local options or API-driven choices like OpenAI and Gemini—to provide intelligent results with similarity scores. Because it follows the Model Context Protocol, it serves as a standardized interface for LLMs to programmatically access metadata and attachments, making it an essential component for building sophisticated, research-aware AI workflows.
Category: AI Memory & Context
Tags: academic, bibliography, citations, research, semantic search, zotero
bash uv tool install "git+https://github.com/54yyyu/zotero-mcp.git" zotero-mcp setup Via pip: bash pip install git+https://github.com/54yyyu/zotero-mcp.git zotero-mcp setup Via Smithery (for Claude Desktop): bash npx -y @smithery/cli install @54yyyu/zotero-mcp --client claude ---bash zotero-mcp setupclaude_desktop_config.json: json { "mcpServers": { "zotero": { "command": "zotero-mcp", "env": { "ZOTERO_LOCAL": "true" } } } }json { "mcpServers": { "zotero": { "name": "zotero", "type": "stdio", "isActive": true, "command": "zotero-mcp", "args": [], "env": { "ZOTERO_LOCAL": "true" } } } }bash zotero-mcp setup --no-local --api-key YOUR_API_KEY --library-id YOUR_LIBRARY_ID ---zotero_semantic_search: AI-powered similarity search with embedding models. * zotero_update_search_database: Manually update the semantic search database. * zotero_get_search_database_status: Check database status and configuration.zotero_search_items: Search library by keywords. * zotero_advanced_search: Perform complex searches with multiple criteria. * zotero_get_collections: List collections. * zotero_get_collection_items: Get items in a collection. * zotero_get_tags: List all tags. * zotero_get_recent: Get recently added items. * zotero_search_by_tag: Search library using custom tag filters.zotero_get_item_metadata: Get detailed metadata (supports BibTeX export via format="bibtex"). * zotero_get_item_fulltext: Get full text content. * zotero_get_item_children: Get attachments and notes.zotero_get_annotations: Get annotations (including direct PDF extraction). * zotero_get_notes: Retrieve notes from your Zotero library. * zotero_search_notes: Search in notes and annotations. * zotero_create_note: Create a new note for an item (beta). ---LaTeX editors (like Cursor or VS Code) often have to leave their environment, open Zotero, find a paper, and manually export the BibTeX citation. Solution: This MCP allows users to retrieve BibTeX metadata directly within their AI-powered IDE or chat interface. You can find the paper and get the correctly formatted citation without breaking your flow. Example: While writing a README in Cursor, the developer asks: "Find the paper I added yesterday about 'Fast Fourier Transforms' and give me its BibTeX citation." The AI returns the BibTeX block ready to be pasted into a .bib file.
Part of MCP Servers
You can install the server via uv using uv tool install zotero-mcp-server or through standard pip. Afterwards, execute zotero-mcp setup to automatically configure connection parameters for supported clients like Claude Desktop. If you need semantic search or PDF features, install the extras by specifying zotero-mcp-server[all].
The server enables AI assistants to search your Zotero library using keywords or semantic embeddings, read full-text papers, retrieve BibTeX records, and extract PDF annotations. It can also create collections, add new papers by DOI or URL, manage tags, and fetch Scite citation metrics or retraction notices.
Zotero MCP works with any client supporting the Model Context Protocol, including Claude Desktop, ChatGPT, Cursor, Cherry Studio, and Chorus. For developer agents with shell access, such as Claude Code or Cursor, it can also run as an agent skill via the standalone zotero-cli utility.
No, an internet connection is not strictly necessary. Zotero MCP supports a fully local operating mode using your local Zotero 7 SQLite database. In this mode, reading metadata and searching local files requires no web API key or active internet connection.