Documentation

RefBridgeMCP connects your Zotero reference library to Claude and other AI assistants over the Model Context Protocol (MCP). Once connected, your assistant can search your library, read attached PDFs, format bibliographies, add and organize references, and — on paid plans — run semantic search and explore your citation network, all through normal conversation.

1. Connecting from Claude

1.1 Claude (web, desktop, and mobile)

1. Open Settings → Connectors → Add custom connector.
2. Enter the server URL: https://refbridgemcp.com/mcp
3. Click Add, then Connect. A RefBridgeMCP window opens — sign in with your Zotero account and approve access. That's the whole setup: no API keys to copy.
4. In a new chat, enable the connector from the tools menu and try: “Search my Zotero library for papers about coastal salinization.”

If RefBridgeMCP is available in your Claude connectors directory, you can skip the URL and just click Connect on the listing — the sign-in flow is identical.

1.2 Other MCP clients

RefBridgeMCP is a standard remote MCP server (streamable HTTP) with OAuth 2.1 and dynamic client registration, so any MCP client that supports remote servers with OAuth can connect using the same URL: https://refbridgemcp.com/mcp. Your dashboard shows the connected clients and lets you revoke any of them at any time.

2. What your assistant can do

Search & read (Free)

Full-text search across your library (search_library), item metadata and counts, item notes and child attachments, extracted PDF text (get_attachment_text), collections, tags, saved searches, and trash listings.

Add & edit (Personal and Pro)

Add references by DOI, ISBN, or URL (add_item_by_doi and friends), create and update items and notes, manage tags and collections, trash/restore/delete items, link related items, download attachment files, fetch open-access PDFs for items that lack one (fetch_pdf_for_item), and repair missing DOIs (heal_missing_dois).

Citations & bibliographies (Personal and Pro)

Format bibliographies in APA, MLA, Chicago, and thousands of other CSL styles (format_bibliography), and export citations as BibTeX, RIS, or CSL-JSON (export_citations).

Semantic search (Personal and Pro)

Meaning-based search over your library's titles, abstracts, notes, and PDF full text (semantic_search) — finds conceptually related work even when the words differ. Requires a one-time index build; see Indexing prerequisites.

Citation intelligence (Pro)

Web-of-Science-style discovery backed by OpenAlex: what a paper cites (get_references), what cites it (get_citing_works), retraction checks (check_retractions), and whole-collection integrity scans (scan_collection_integrity).

Knowledge graph (Pro)

An entity graph of the works, authors, venues, and concepts in your library: search and cross-reference entities, find co-author networks (find_coauthors), traverse citation chains (traverse_citations), and open an interactive graph view in your browser (open_graph_view). Citation arrows point from the older, cited work to the newer, citing work — the direction knowledge flows.

A safety net for every edit

Every write your assistant makes is recorded in an audit log you can review (list_recent_edits) and reverse (undo_library_edit). Items touched by the assistant are tagged ai-review so you can find them in Zotero.

3. Plans

Daily and per-message limits are adjustable from your dashboard up to your plan's ceiling. Full details on the pricing page.

4. Indexing prerequisites

Two features need a one-time index before first use. Your assistant can run both.

• Semantic search — ask your assistant to run reindex_library (repeating until it reports done: true). It indexes titles, abstracts, notes, and PDF text; PDFs whose text Zotero hasn't extracted yet are tracked and picked up automatically on later runs. New and edited items are indexed incrementally afterward. Check progress any time with get_index_status.
• Citation intelligence and the knowledge graph — ask your assistant to run reindex_entities (also repeating until done: true), which enriches your library against OpenAlex. Only needed if graph or citation tools report an empty index (get_entity_index_status shows coverage).

Retraction checks and citation lookups work through DOIs — if older entries in your library lack them, heal_missing_dois fills them in from CrossRef first.

5. Prompts to try

• “Find the papers in my library about groundwater-driven salinization and summarize what they disagree on.”
• “Add this paper to my library: 10.1029/2005wr004302 — then fetch an open-access PDF for it if one exists.”
• “Format an APA bibliography of everything in my ‘Methods’ collection.”
• “Check my library for retracted papers.” (Pro)
• “Open a citation graph around Wilson & Gardner 2006 and show me what built on it.” (Pro)
• “What did you change in my library this week? Undo the tag edits.”

6. Troubleshooting

• Semantic search returns nothing. The index probably hasn't been built — run reindex_library until done: true (Personal/Pro).
• Graph or citation tools return an empty result. Run reindex_entities until it reports done: true, then retry.
• “Item does not exist.” Your connector may be signed into a different Zotero account than you expect. Reconnect from Claude's connector settings and check the account shown during sign-in.
• Rate-limit messages. Each message names the limit that was hit (daily, per-message, or monthly). Daily and per-message caps are adjustable on your dashboard; monthly caps are fixed per plan.
• The graph view asks me to sign in. That's intentional — graph sessions are private to your RefBridge account. Sign in with Zotero and you'll land back on your graph.

7. Security & privacy

• Authentication is OAuth end to end — you never handle API keys, and you can revoke any connected client from your dashboard.
• Serving a request never stores your library content beyond short-lived caches. The indexing features you opt into do persist derived data: semantic search stores embeddings and the text chunks they were computed from, and the entity graph stores work/author/ venue/concept rows derived from your library. Turning embeddings off in your dashboard removes the semantic index; details in the Privacy Policy.
• Every AI write is recorded in the audit log. Most are undoable with undo_library_edit; permanent deletions and library-wide operations (tag renames/deletes, collection and saved-search changes) are logged for visibility but cannot be automatically reversed.
• Details in the Privacy Policy and Terms of Service.

8. Support

Questions or problems: support@personailized.com. RefBridgeMCP is not affiliated with, sponsored by, or endorsed by the Corporation for Digital Scholarship or the Zotero project.