Zotero MCP connects AI assistants to a local Zotero reference library through the Model Context Protocol (MCP). It ships as a plugin for Zotero 7 with the MCP server built in, so an assistant talks to Zotero over Streamable HTTP and no separate server process runs alongside it.
The integrated server exposes 20 tools in five groups: search and query, collection management, semantic search, a full-text database, and write operations. Searches cover titles, creators, years, tags, item types, annotations, and cached PDF full text, with boolean operators, relevance scoring, and four levels of returned detail (minimal, preview, standard, complete).
Semantic search indexes items as embedding vectors stored with SQLite-vec and matches on concepts rather than keywords, including across languages; embeddings come from an OpenAI or Ollama endpoint, detected automatically. Write tools create and update notes, tags, metadata fields, and items, reparent standalone PDFs, and import records by DOI, arXiv ID, ISBN, PMID, or ADS bibcode through Zotero's own resolvers. Semantic search and write operations can each be switched off in preferences.
Setup is a plugin install plus enabling the server in Zotero preferences, which generates the connection configuration to paste into a client. Documented clients include Claude Desktop, Cherry Studio, and Cursor. The server listens on port 23120 by default and operates only against the local library.
Features
- Integrated MCP server: the Zotero plugin serves MCP over Streamable HTTP without a second process
- Advanced search: boolean operators, relevance scoring, and filters on title, creator, year, tags, and item type
- Content extraction: PDF full text, attachments, notes, abstracts, and webpage snapshots at four detail levels
- Annotation search: highlights, notes, and annotations retrieved by colour, tag, or keyword with ranking
- Collection browsing: collection hierarchies, subcollections, details, and item lists
- Semantic search: embedding-vector matching with SQLite-vec indexing and an index-status column in the library view
- Write operations: notes, tags, metadata fields, new items, and reparenting of standalone attachments
- Identifier import: DOI, arXiv, ISBN, PMID, and ADS bibcode lookups using Zotero's native translators
- Full-text database: list, search, get, and stats over the cached PDF full-text store
- Client configuration generator: builds the connection block for a given assistant from the preferences pane
