Zotero MCP Server: Connect AI to Papers, Notes, and Citations

A Zotero MCP server lets an AI assistant search your Zotero library, read item metadata, access notes, and retrieve citation details during a research or writing session. Instead of copying references into a chat window by hand, the assistant queries Zotero directly and uses the results in its answers. Tools like fenbs keep research questions, source links, and conclusions in one place so the context carries across AI apps and sessions.

Zotero is a free reference manager that stores papers, books, articles, and notes with full metadata. An MCP connection makes that library a live resource for your AI assistant rather than a separate application to check manually.

What a Zotero MCP server can do

The Zotero API covers items, collections, tags, notes, and attachments. An MCP server wraps those endpoints as tools the assistant can call. Common capabilities include searching items by keyword or tag, reading full metadata for a source, listing items in a collection, and retrieving notes attached to a reference.

• Item search: find references by title, author, year, tag, or keyword across your library.
• Metadata read: retrieve the full citation details for a specific item, including abstract, DOI, and URL.
• Collection browse: list items in a named collection, useful when research is organized by project.
• Notes access: read notes attached to a reference without opening the Zotero desktop app.

Check the server's documentation for which API methods it exposes and whether it connects to a local Zotero instance or to the Zotero cloud API. The cloud API requires a Zotero account and an API key from your account settings.

Finding and evaluating a Zotero MCP server

Search the MCP server directory for Zotero integrations. When evaluating one, check whether it connects to the Zotero cloud, a local storage instance, or both. Also check whether it supports group libraries if your research is shared with colleagues.

The Zotero API documentation on the Zotero website describes available endpoints and rate limits. A server that exposes read-only access is a safe starting point; adding write tools (creating items or notes) makes sense only if the assistant will help build the library, not just query it.

Checking citations the assistant retrieves

An MCP connection gives the assistant access to the metadata you have already saved in Zotero. It does not verify that a DOI resolves to the right paper or that a URL still works. For critical citations, open the primary source and confirm the claim the assistant attributes to it.

Tag items in Zotero when you have verified them against the primary document. Ask the assistant to note whether a source is tagged as verified or unverified. That small distinction prevents a search result from being treated as confirmed evidence in a draft.

Keeping research links consistent across AI apps

Research projects often span multiple AI sessions and more than one AI app. A question you answered in one chat may be the foundation for a later analysis in a different tool. Without a shared record, that earlier conclusion has to be reconstructed or explained again.

A shared links record stores the sources the team has used, what each one was used for, and whether it was checked. Alongside that, a context note saves the research question, the scope decisions, and the current open threads so a fresh assistant can pick up the work.

Setting context at the start of a writing session

Before asking an assistant to draft or analyze anything, ask it to read the project context: the research question, the key sources, any scope decisions, and where the previous session ended. That startup step replaces the paragraph of explanation you would otherwise type at the beginning of each new chat.

With a Zotero MCP connection active, the assistant can then pull specific items by tag or collection name rather than working from memory. "Find all items tagged 'primary-source' in the Urban Planning collection" is more precise than "search for relevant papers." For broader search and sourcing strategies, see best MCP servers and mcp-server-configuration.

Tag-based queries work best when your Zotero library uses consistent tags. Agree on a tagging convention before connecting: decide which tags mark primary sources, verified items, and items still under review, and apply them consistently. An assistant that queries by tag is only as useful as the tagging behind it.

How to do this with fenbs

1. Create a free account and add a project for your research area.
2. Add Links and listings for the key sources: the Zotero item URL or DOI, the purpose of the source, and whether you have verified it against the primary document.
3. Connect your AI assistant over MCP from Integrations. At the start of each session, ask it to read the project context and the current source list before drafting or analyzing.
4. Record scope decisions in Decisions and rules: for example, which databases are in scope, which time periods to exclude, or which types of sources count as primary.
5. Before ending a session, save where the research stands and what threads are still open so the next session starts from that point rather than asking you to re-explain the scope.