Zotero MCP Server

The Zotero MCP Server is a lightweight, prototype service that bridges the gap between an AI assistant and a Zotero reference library. By exposing search and retrieval capabilities through the Model Context Protocol, it lets an assistant query a user’s entire Zotero collection—spanning PDFs, notes, and other item types—and then fetch the full content of selected items. This removes the need for manual export or complex API handling, enabling a seamless research workflow directly inside conversational agents.

For developers building AI‑powered research tools, the server offers a concrete example of how to expose domain‑specific data sources via MCP. The server is built on the popular Pyzotero client, which handles authentication and request formatting for Zotero’s REST API. Once the server is running, a client can invoke two core tools: search_zotero_library and retrieve_zotero_item_content. The search tool accepts a query string and an optional limit, returning metadata such as title, type, parent collection, and the unique Zotero key. The retrieve tool takes a list of those keys and returns the parsed content (text, PDFs, or notes) along with additional metadata. This separation of concerns mirrors typical AI workflows—first locate relevant sources, then ingest and summarize them.

Key capabilities include:

• Full‑library search: Query across all collections, authors, tags, and notes with a single request.
• Structured results: Metadata is returned in a machine‑readable format, enabling the assistant to present concise tables or lists.
• Content extraction: The server fetches raw text from PDFs and notes, which can then be fed to a summarization or question‑answering model.
• Scalable limits: The search function accepts a parameter, allowing the client to request only as many results as needed while still supporting large queries (up to 100 items).

Typical use cases span academic research, literature reviews, and knowledge management. A researcher could ask an assistant to “find recent papers on deep learning in medical imaging” and receive a curated list of Zotero items. The assistant can then request the full text of selected papers, summarize key findings, or even extract citations for a bibliography. In corporate settings, teams can quickly surface internal whitepapers or meeting notes stored in Zotero without leaving their chat platform.

Integration is straightforward: developers can add the server as a tool in any MCP‑compatible client (the README references the 5ire chat interface). Once registered, the assistant can invoke the tools via system prompts or tool calls. Because the server operates over HTTP and follows MCP conventions, it can be hosted behind a reverse proxy or within a container, making it adaptable to both local and cloud environments. The prototype demonstrates how domain‑specific data sources can be wrapped in a standard protocol, providing a reusable pattern for future integrations.