zotero-local-mcp
Zotero Local MCP is a local-first MCP server + CLI that lets AI agents search, read, annotate, manage, and organize a local Zotero library via the Zotero 10 local API.
Search & discovery: keyword/title/creator/year search, advanced structured-field search, tag search, citation-key lookup, and semantic/topic search across the active or all libraries.
Reading: retrieve full-text content, read specific PDF page ranges, extract PDF outlines, get local attachment paths, and list child attachments/notes.
Annotations & notes: create text highlights and area annotations on PDFs, update/trash annotations, get/synthesize annotations, and create/update/trash Markdown/HTML notes.
Item management: add items by DOI, arXiv, URL, ISBN, BibTeX, CSL JSON, or local file; attach files to existing items; update metadata, creators, tags, and collections; move items to the trash (recoverable).
Collections: list, search, create, and delete collections; assign items to collections; page through collection contents.
Library operations: list and switch libraries, view recently added items, batch-update tags/Extra fields, and export formatted bibliographies (APA, MLA, Chicago, IEEE, BibTeX).
Local-first features: local authorization without API keys, files stored directly in Zotero local storage, reversible trash-based deletes, WebDAV sync handled by Zotero desktop, and a
zotero-clifor scripted use.
Supports importing references by arXiv identifier, adding metadata and optionally downloading open-access PDF attachments to local storage.
Supports importing references by DOI, automatically adding metadata and downloading available open-access PDFs to the local Zotero storage.
Provides tools for searching, reading, annotating, and managing a local Zotero library, including items, collections, notes, attachments, and safe deletion to the trash.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@zotero-local-mcp Find papers about resource dependence theory in my Zotero library."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
项目介绍
Zotero Local MCP 基于 Zotero 10 本地 API(http://localhost:23119/api/)构建,提供 Model Context Protocol (MCP) 服务与配套命令行工具 (zotero-cli)。
支持 Claude Desktop、Cursor、Oh My Pi (OMP) 等 AI 客户端直接检索、读取、批注与管理本地 Zotero 文献库。附件在本地文件系统直接落盘,不经由 Zotero 官方云端存储转发。
Related MCP server: zotero-mcp
架构特性
1. 本地落盘与存储兼容
本地存储:通过 DOI、arXiv、URL 或本地路径添加的 PDF 附件直接复制至本地
storage/<key>/目录,不向 Zotero 云端上传文件。WebDAV 同步:条目与文件写入本地后,由 Zotero 桌面端按自身设置执行 WebDAV 同步,MCP 不介入网络同步链路。
2. 本地授权与凭据安全
基于 Zotero 10
POST /api/local/authorize规范握手。授权 Token 存储于本地
~/.config/pyzotero/local-api-key.json(文件权限0600)。客户端配置无需显式填写明文 API Key。
3. 可逆删除
文献、笔记与注释的删除通过
PATCH {"deleted": 1}移至 Zotero 回收站,可在桌面客户端中手动恢复,不执行物理删除。
4. 环回代理保护
服务端绑定
NO_PROXY=127.0.0.1,localhost,防止本机系统代理导致本地 API 请求被误路由。
5. 双接口支持
MCP 服务:提供 30+ 项工具,供 AI 客户端调用。
命令行工具 (
zotero-cli):支持无 AI 环境下的脚本批处理,提供--json输出。
功能清单
模块 | 核心能力 | MCP 工具 / CLI 命令 |
文献检索 | 关键词检索、多字段组合检索、跨文库检索、标签检索 |
|
论文阅读 | PDF/EPUB 文本提取、目录书签解析、指定页码读取 |
|
批注与笔记 | PDF 文本高亮、图形区域框选标注、Markdown/HTML 笔记读写 |
|
文献录入 | 基于 DOI、arXiv、URL、ISBN、BibTeX 或本地文件导入 |
|
集合整理 | 集合创建与删除、条目分类转移、批量标签更新 |
|
条目维护 | 条目移入回收站、元数据更新、重复项排查 |
|
安装与配置
1. 前提要求
安装 Zotero 10+。
在 Zotero 设置中允许本机通信:
macOS:
Settings→Advanced→ 勾选 Allow other applications on this computer to communicate with Zotero。Windows/Linux:
Edit→Preferences→Advanced→ 勾选相同选项。
2. 安装
使用 uv 安装:
# 从源码安装
git clone https://github.com/JingYangYuan/zotero-local-mcp.git
cd zotero-local-mcp
uv tool install .
# 或直接通过 pip 安装
pip install .3. 本地授权
确保 Zotero 桌面端正在运行,执行本地授权:
pyzotero authorize --app-name "Zotero MCP Local"Zotero 桌面端将弹出确认窗口,选择 Always Allow。凭据将自动保存至本地。
检查状态:
zotero-cli --json config输出 ZOTERO_LOCAL: true 且无错误即表示配置完成。
客户端配置
1. Oh My Pi (OMP)
在 ~/.omp/agent/mcp.json 中添加:
{
"mcpServers": {
"zotero": {
"type": "stdio",
"command": "zotero-mcp-server",
"args": ["serve"],
"env": {
"ZOTERO_LOCAL": "true",
"ZOTERO_MCP_SCHEMA_REFRESH": "0"
}
}
}
}2. Claude Desktop
在 Claude Desktop 配置文件(macOS: ~/Library/Application Support/Claude/claude_desktop_config.json,Windows: %APPDATA%\Claude\claude_desktop_config.json)中添加:
{
"mcpServers": {
"zotero": {
"command": "zotero-mcp-server",
"args": ["serve"],
"env": {
"ZOTERO_LOCAL": "true",
"ZOTERO_MCP_SCHEMA_REFRESH": "0"
}
}
}
}3. Cursor
在 .cursor/mcp.json 中添加:
{
"mcpServers": {
"zotero": {
"command": "zotero-mcp-server",
"args": ["serve"],
"env": {
"ZOTERO_LOCAL": "true",
"ZOTERO_MCP_SCHEMA_REFRESH": "0"
}
}
}
}命令行 (zotero-cli) 使用
# 1. 检索文献
zotero-cli search "关键词"
zotero-cli --json search "关键词" --limit 5
# 2. 查看元数据与子附件
zotero-cli get metadata <ITEM_KEY>
zotero-cli get children <ITEM_KEY>
# 3. 添加文献与附件
zotero-cli add doi 10.1145/3708319
zotero-cli add file --filepath /path/to/paper.pdf --title "论文标题"
# 4. 集合管理
zotero-cli collections search "分类名"
zotero-cli collections manage --item-keys <ITEM_KEY> --add-to <COLL_KEY_A> --remove-from <COLL_KEY_B>
# 5. 移入回收站
zotero-cli delete item <ITEM_KEY>常见问题
Q: 是否需要配置 ZOTERO_API_KEY 或 ZOTERO_LIBRARY_ID?
A: 不需要。Zotero 10 本地 API 支持本地授权与写入,读写操作均在本地执行,无需云端 API Key。
Q: 本地添加的文献与附件是否会自动同步到其他设备?
A: 会。只要在 Zotero 桌面端配置了 WebDAV 或官方同步,附件写入本地 storage/ 后,桌面端会按既有规则自动同步。
Q: 误删条目如何恢复?
A: 打开 Zotero 桌面端,在左侧侧边栏进入“回收站”,右键目标条目选择“恢复到文库”。
许可证
本项目基于 MIT License 开源。
Available Tools
38 toolszotero_add_itemA
Add item(s) to Zotero from any source: DOI, URL, ISBN, BibTeX, CSL JSON, or a local file. Use for every 'add this to Zotero' request. source: the identifier, URL, citation text, or ABSOLUTE file path. DOI/URL/ISBN also take many at once (list or comma/newline-separated), each resolved independently. BibTeX/CSL JSON may be inline (many entries per call) or a path to .bib/.bibtex/.json/.csljson; documents are .pdf, .epub, .docx and similar. source_type: 'auto' (default) detects it, incl. comma/newline DOI lists; override for URL/ISBN batches. Routing: doi → CrossRef (best metadata — prefer a DOI when you have one); url → doi.org/arxiv.org get full metadata, anything else becomes a bare 'webpage' item that is often not citable, so resolve to a DOI first; isbn → Open Library then Google Books (noisy — verify after); bibtex/csl_json → one item per entry, citation key kept in Extra; file → extracts the PDF's DOI and enriches via CrossRef, else guesses from filename/text, then attaches the file. collections: keys, names, or '/'-paths ('_project/topic'), validated before anything is created — an unknown or ambiguous spec fails the call rather than leaving an unfiled item; create_missing_collections=True creates them instead. if_exists: 'duplicate' (default) always creates; 'file' is idempotent — reuses the item matching the DOI/ISBN/URL, adding missing collections/tags, never removing; 'skip' leaves a match untouched. attach_mode: 'auto' (default) attaches an OA PDF, 'linked_url' bookmarks it, 'none' skips, 'required' fails without one. title: file sources only, when extraction misses. Requires local Zotero write authorization. Attached files are copied into local Zotero storage; subsequent syncing is determined by Zotero desktop settings. Run Example: zotero_add_item(source='10.1145/3708319', collections=['9SU943GB'], if_exists='file').
Inputs: source (required), tags, title, if_exists, +4 more. Returns: result
zotero_advanced_searchA
Advanced item search with multiple structured-field conditions joined by AND or OR. Use this when you need to filter by fields that zotero_search_items and zotero_search_by_tag can't express (date ranges, specific itemTypes, etc.). For plain text use zotero_search_items; for tags use zotero_search_by_tag; for topic discovery use zotero_semantic_search. conditions: list of {field, operation, value} dicts (also accepts a JSON string). Common fields: title, creator, date, dateAdded, dateModified, tag, itemType, publicationTitle, abstractNote, collection. Supported operations (exhaustive): is, isNot, contains, doesNotContain, beginsWith, endsWith, isGreaterThan, isLessThan, isBefore, isAfter. For 'added in the last N days', use field='dateAdded' with operation='isAfter' and an ISO date value (e.g. '2026-03-22'). join_mode: 'all' (AND, default) or 'any' (OR). sort_by: dateAdded, dateModified, title, creator, etc. sort_direction: 'asc' (default) or 'desc'. limit: max results (default 50, max 500). include_subcollections: make a 'collection' condition match items anywhere in that collection's subtree, for the is/isNot operations (default False). search_all_libraries: search every accessible library at once, labelling each result with its library; needs ZOTERO_SEARCH_BACKEND=sqlite. 'tag' conditions work; 'collection' conditions and include_subcollections do not. Example: zotero_advanced_search(conditions=[{'field': 'itemType', 'operation': 'is', 'value': 'preprint'}, {'field': 'dateAdded', 'operation': 'isAfter', 'value': '2026-03-22'}], join_mode='all').
Inputs: conditions (required), limit, sort_by, join_mode, +3 more. Returns: result
zotero_attach_fileA
Attach a file to an EXISTING Zotero item as an imported child attachment (uploads the file bytes). Use when the item is already in the library and you have its key — e.g. attaching a PDF you found for a reference. To create a NEW item from a file, use zotero_add_from_file instead. item_key: key of the existing REGULAR item. Passing an attachment/note key fails with a hint to use its parent. file_path: ABSOLUTE local path (.pdf, .epub, .djvu, .doc, .docx, .odt, .rtf). url: direct http(s) link, downloaded server-side — PDF-only; for other formats download locally and use file_path. Exactly one of file_path/url must be given. filename: optional stored-filename override; defaults to the file's basename or the URL's last path segment (falling back to <item_key>.pdf); a missing extension is appended automatically. Returns the created attachment's key. Idempotent: if the item already has an attachment with the same filename or identical content (MD5), nothing is re-uploaded. Requires local Zotero write authorization. Files are copied into local Zotero storage; subsequent syncing is determined by Zotero desktop settings. Run zotero_update_search_database afterwards to index the new file for semantic search. Example: zotero_attach_file(item_key='ABCD2345', file_path='/Users/me/smith-2020.pdf').
Inputs: item_key (required), url, filename, file_path. Returns: result
zotero_batch_updateA
Edit metadata across many items in one call: add/remove tags and upsert/remove Key: value lines in Extra (Better BibTeX keys, tex.* fields). Select items by item_keys, and/or a free-text query, and/or an existing tag (query and tag are ANDed; tag may be a list to OR); item_keys wins. At least one selector AND one action are required. add_tags/remove_tags keep the item's other tags — not a replace-all. set_keys upserts Extra lines, matching a line case-insensitively by its key: prefix and replacing it in place, else appending; remove_keys deletes those lines; lines without a colon are preserved. limit: max items for query/tag selection (default 50). Attachments and items needing no change are skipped and counted. Requires a writable library. Example: zotero_batch_update(tag='to-read', add_tags=['reviewed'], remove_tags=['to-read']).
Inputs: tag, limit, query, add_tags, +4 more. Returns: result
zotero_create_annotationA
Create an annotation on a PDF attachment (EPUB: highlights only). Exactly one of two modes per call: text= HIGHLIGHTS selectable text; rect= draws an AREA box over a figure, table, or other non-text region (PDF only). Passing both or neither is an error. attachment_key: the PDF/EPUB attachment key, NOT the parent item key (zotero_get_item_children finds it). page: 1-indexed page (EPUB: 1-indexed chapter). text: exact text to highlight, matched against the text layer — scanned/image-only PDFs will not match. rect: [x, y, width, height] normalized to [0, 1], with (0, 0) at the page's top-left; width/height are page-relative and the box must fit the page. Call zotero_get_page_layout first and reuse a detected region's bbox instead of guessing coordinates. comment, color (hex, default '#ffd400'), tags: optional. Requires PyMuPDF (the [pdf] extra) and a writable library (web API key or hybrid mode). Examples: (attachment_key='NHZFE5A7', page=4, text='working memory'); (attachment_key='NHZFE5A7', page=7, rect=[0.15, 0.22, 0.6, 0.35], comment='Figure 3').
Inputs: page (required), attachment_key (required), rect, tags, +3 more. Returns: result
zotero_create_collectionA
Create a new collection (project/folder) in your Zotero library. To create a subcollection, pass parent_collection (not parent_key) as either a collection key (8-character string like 'KMMQDFQ4') or a collection name. Use zotero_search_collections to find collection keys.
Inputs: name (required), parent_collection. Returns: result
zotero_delete_annotationA
Move a Zotero annotation to the Trash. Trashed annotations are recoverable from Zotero's Trash — empty the Trash in the Zotero UI for permanent deletion.
Inputs: annotation_key (required). Returns: result
zotero_delete_collectionA
Delete a collection (folder) from your Zotero library by its 8-character key. Items inside the collection are NOT deleted — they remain in the library (and in any other collections they belong to). Subcollections ARE deleted along with the parent. This is a hard delete — Zotero's API does not trash collections, so the operation cannot be undone via the API. Use zotero_search_collections to find the key first. Example: zotero_delete_collection(collection_key="KMMQDFQ4").
Inputs: collection_key (required). Returns: result
zotero_delete_itemA
Move a Zotero item to the Trash. Works for any item type (book, journalArticle, webpage, attachment, etc.). For notes, use zotero_delete_note — identical mechanism, constrained to notes for safety. Trashed items are recoverable from Zotero's Trash — empty the Trash in the Zotero UI for permanent deletion. By default refuses to trash notes; set allow_note=True to override.
Inputs: item_key (required), allow_note. Returns: result
zotero_export_bibliographyA
Render a formatted bibliography or in-text citations for a set of Zotero items using Zotero's own CSL citation engine, so you can drop references straight into a manuscript. item_keys: optional list of 8-character item keys (also accepts a JSON list string); takes precedence over collection_key. collection_key: optional collection to export instead; if neither is given, the active library is exported (capped). style: CSL style short name (default 'apa'); e.g. 'modern-language-association', 'chicago-note-bibliography', 'ieee'. Ignored for bibtex. export_format: 'bib' (formatted reference-list entries, default), 'citation' (in-text citation strings), or 'bibtex' (raw BibTeX for .bib files). Output: markdown naming the style/format, then the rendered entries (a fenced block for bibtex, a numbered list otherwise). Rendering uses Zotero's own CSL engine and works in local mode with no API credentials, as well as over the web API. Capped at 100 items per call; scope with item_keys or collection_key for anything larger. Example: zotero_export_bibliography(item_keys=['RTKZQI8E'], style='apa', export_format='bib').
Inputs: style, item_keys, export_format, collection_key. Returns: result
zotero_get_annotationsA
Get annotations (highlights and attached notes on PDF/EPUB attachments) for a specific item or across the active Zotero library. item_key: pass the parent item key OR an attachment key — both work; attachment-to-parent resolution is automatic. ALWAYS pass item_key when you know which item you want; calling without it returns every annotation in the library (potentially thousands). use_pdf_extraction=True falls back to direct PDF parsing when the Zotero API has no stored annotation record — useful for annotations made outside Zotero desktop. limit: cap on annotations returned; None (default) returns all. format='markdown' (default) returns a readable list; format='json' returns normalized records with stable keys for downstream scripts and other MCP tools. Uses Better BibTeX when Zotero desktop is running locally, otherwise the Zotero web API. Example: zotero_get_annotations(item_key='ABC12345') → every highlight/note on that paper.
Inputs: limit, format, item_key, use_pdf_extraction. Returns: result
zotero_get_attachment_pathA
Return the local filesystem path(s) of a Zotero item's attachments. Local mode only. Useful when you want to read a large PDF directly (e.g., a book) instead of going through zotero_get_item_fulltext, which is page-limited.
Inputs: item_key (required). Returns: result
zotero_get_collection_itemsA
Get all items in a specific Zotero collection. Supports detail='keys_only' (minimal), 'summary' (default, no abstracts), or 'full' (with abstracts). Includes PDF/notes indicators. include_subcollections=True also returns items filed in collections nested beneath this one (default False, matching Zotero's own 'Search subcollections' checkbox). For a collection larger than limit, page through it with offset (the response names the next offset to pass). TIP: To find papers on a specific topic, use zotero_semantic_search instead — it's faster and returns only relevant results.
Inputs: collection_key (required), limit, detail, offset, +1 more. Returns: result
zotero_get_collectionsA
List all collections in the currently active Zotero library as a hierarchical tree (parents and nested subcollections, each with its 8-character key). Use this when the user wants to see the full library structure. If you already know a name and just need the key, prefer zotero_search_collections — it returns only matches. Scope is limited to the active library — switch libraries with zotero_switch_library before listing. Deep hierarchies render inline without truncation, so very deep trees can be long. limit: cap on collections returned; pass None (default) to use 100, or raise to 5000 for libraries with thousands of collections. include_trashed: when True, also show collections in the Zotero Trash (annotated as such). Default False, matching Zotero desktop's default view. Example output:
Orals (Key: MT53KB66)
Early America (Key: 3249BZKE)
I. Historiography & Methodology (Key: XFN79DUT)
Inputs: limit, include_trashed. Returns: result
zotero_get_item_childrenA
List the child items (attachments, notes, annotations under an attachment) of one OR MANY parent Zotero items. Use it to find an item's PDF/EPUB attachment key before zotero_create_annotation or zotero_get_pdf_outline — those take an attachment key, NOT the parent item key. item_key: one 8-character parent key, or an ARRAY of keys (a JSON-encoded list string also works). Pass every key you have in ONE call: a batch is one API round trip instead of N, and a bad key is reported in its own section instead of aborting. Returns markdown — one key: attachments (content type, filename) and notes in full under the parent title; several keys: one compact line per child, grouped under each parent. Scope: active library only. Examples: zotero_get_item_children(item_key='RTKZQI8E'); zotero_get_item_children(item_key=['RTKZQI8E', '9UZR8GXT']).
Inputs: item_key (required). Returns: result
zotero_get_item_fulltextA
Return the full extracted text of a Zotero item's primary attachment (PDF or EPUB). WARNING: returns the entire paper (often 10K+ tokens). Use ONLY when the user explicitly wants to READ the paper — not for searching or browsing. For topic search use zotero_semantic_search; for metadata only use zotero_get_item_metadata. Avoid calling this on multiple papers in one conversation unless the user specifically asked to read several. item_key: 8-character Zotero item key. Normally the parent item — the tool locates the attached PDF/EPUB itself, preferring PDF unless attachment_priority says otherwise. Passing an attachment's own key instead reads exactly that file and skips the priority order, which is how you read one specific attachment of an item that has several (find keys via zotero_get_item_children). Scope: active library only. Extraction path (in order): local Zotero storage via SQLite when running in local mode (fastest, respects pdf_max_pages config); Zotero's server-side fulltext index; direct download and parsing as a last resort. Image-only scanned PDFs without OCR may return little or no text. Example: zotero_get_item_fulltext(item_key='RTKZQI8E').
Inputs: item_key (required). Returns: result
zotero_get_item_metadataA
Fetch detailed metadata (title, creators, date, DOI, publisher, tags, abstract, URL, etc.) for ONE Zotero item by key. If the metadata and abstract don't contain what you need, call zotero_get_item_fulltext to read the paper — but that is resource-intensive (10K+ tokens) and should NEVER be used for searching; use zotero_search_items or zotero_semantic_search instead. item_key: the 8-character Zotero item key (NOT a DOI or title). include_abstract=True (default) includes the abstractNote in markdown output; pass False to trim tokens when you don't need it. (Ignored in bibtex/json formats.) format='markdown' (default) returns a human-readable block; format='json' returns the complete raw Zotero item record; format='bibtex' returns a BibTeX citation string suitable for .bib files. Scope: active library only (switch with zotero_switch_library). Unlike list endpoints, this returns items EVEN IF THEY ARE IN THE TRASH — a Status: In Trash line is surfaced when the item is trashed (recoverable via the Zotero UI). Collection membership is shown as keys rather than a bare count so the caller can verify entries against zotero_search_collections (the Zotero API does not cascade collection-delete to items, so dangling references can linger). Example: zotero_get_item_metadata(item_key='RTKZQI8E', format='bibtex').
Inputs: item_key (required), format, include_abstract. Returns: result
zotero_get_notesA
Read notes from the active Zotero library. Omit query to LIST notes: with item_key, that item's child notes; without it, notes library-wide (capped by limit). Pass query to SEARCH note and annotation text instead — case-insensitive substring over the stripped-text body, library-wide, so query and item_key cannot be combined. limit: max results (default 20). truncate=True (default) shortens long bodies for display; pass False for complete content (list mode only). raw_html=True returns a note's original HTML instead of stripped text — use it when you intend to edit and round-trip via zotero_manage_note(action='update'). Scope: active library only (zotero_switch_library to change). Example: zotero_get_notes(item_key='ABC12345', raw_html=True); zotero_get_notes(query='mindfulness').
Inputs: limit, query, item_key, raw_html, +1 more. Returns: result
zotero_get_page_layoutA
Detect candidate figure/table regions on a PDF page and return their normalized bounding boxes, so area annotations can be placed on detected content instead of guessed positions. ALWAYS call this before zotero_create_annotation's area mode unless exact coordinates are already known. Returns each region's bounding box (x, y, width, height in [0, 1]), source (image/drawing/table/merged), associated caption (e.g. 'Figure 3: ...'), confidence level, and a ready-to-paste zotero_create_annotation call. Note: detection is geometric — boxes cover the graphical core of a figure/table; text labels inside figures or unruled table headers may fall outside the box. Confidence reflects caption matching, not box completeness. attachment_key: PDF attachment key — NOT the parent item key (use zotero_get_item_children to find attachments). page: 1-indexed page number (page 1 is the first page). Scope: PDFs only — EPUB attachments are NOT supported. Read-only: works in both local and web API modes. Example: zotero_get_page_layout(attachment_key='NHZFE5A7', page=7).
Inputs: page (required), attachment_key (required). Returns: result
zotero_get_pdf_outlineA
Extract the table of contents (outline/bookmarks) from a PDF attachment, returned as a hierarchical markdown list with each entry's page number. Use this to orient in a paper before calling zotero_get_item_fulltext — the outline is typically < 200 tokens versus 10K+ for the full text. If the PDF has no embedded outline, returns a short 'no outline' message rather than failing. item_key: the PDF ATTACHMENT key OR the parent item key — both are accepted; attachment-to-parent resolution is automatic. Find the right key with zotero_get_item_children if unsure. Scope: PDFs only (EPUBs have no outline extraction here). Requires PyMuPDF (the [pdf] extra). Read-only; works in local or web mode. Example: zotero_get_pdf_outline(item_key='RTKZQI8E').
Inputs: item_key (required). Returns: result
zotero_get_recentA
List the most recently ADDED items (by dateAdded) in the active library, optionally scoped to a single collection. Use this for 'what did I add recently?' questions — NOT for general topic search (use zotero_semantic_search) or for a collection's full contents (use zotero_get_collection_items). limit: how many recent items to return (default 10). collection_key: optional 8-character collection key to restrict results to that collection; when omitted, returns the N most recent items across the whole library. Ordering is dateAdded DESC. All item types are returned, INCLUDING standalone notes and attachments — so results can mix papers, notes, and loose PDFs. If you only want parent items, filter client-side by itemType in the output. Scope: active library only (switch with zotero_switch_library). Example: zotero_get_recent(limit=20) or zotero_get_recent(collection_key='MT53KB66', limit=5).
Inputs: limit, collection_key. Returns: result
zotero_get_search_database_statusA
Report the semantic search database's readiness and stats: item count, last update time, embedding provider / model, and whether the [semantic] optional dependency is installed. Use this to decide whether zotero_semantic_search will return useful results, or whether the user should run zotero_update_search_database first. Takes no parameters; no side effects. Returns a human-readable status block. If the [semantic] extras are not installed, returns an install hint instead of stats. Example: zotero_get_search_database_status() → count, last sync, provider summary.
Inputs: no named parameters. Returns: result
zotero_get_tagsA
List all tags used in the currently active Zotero library, as a flat markdown list (one tag per line). Use this for tag discovery before filtering with zotero_search_by_tag or batch-editing with zotero_batch_update. Scope is the active library only — switch with zotero_switch_library before listing. The list is flat: tags have no parent/child structure in Zotero, only a colon convention ("area/subtag") that this tool preserves verbatim. limit: cap on tags returned; None (default) returns all. Example output:
to-read
methods/qualitative
AI agents
Inputs: limit. Returns: result
zotero_list_librariesA
List every Zotero library this MCP can address: the user's personal library (libraryID=1 conventionally), all group libraries the user is a member of (with groupID), and (in local mode) RSS feed libraries. Each entry shows the library/group ID, display name, and item count. Use this to discover a library ID before calling zotero_switch_library — the two form a read-then-switch workflow. If the user only wants to see Zotero collections inside the CURRENT library, use zotero_get_collections instead. No parameters. In local mode: reads the local Zotero SQLite DB (fast, includes RSS feeds). In web mode: queries /groups via the Zotero web API (no feeds). Read-only; no side effects. The active library isn't flagged in the output — track it yourself from the last successful zotero_switch_library call (or the ZOTERO_LIBRARY_ID env var if none). Example: zotero_list_libraries().
Inputs: no named parameters. Returns: result
zotero_manage_noteA
Create, update, or trash a Zotero note. item_key: the PARENT item's key for action='create', the NOTE's own key for 'update' and 'delete' (zotero_get_notes finds it). create: needs note_text — plain text, or simple HTML (p, strong, em, ul/li, a, code), which is preserved; note_title becomes a heading; tags optional. update: needs note_text. append=False (default) REPLACES the whole body, append=True concatenates. To keep formatting, fetch with zotero_get_notes(raw_html=True), edit that HTML, and pass it back whole. delete: moves the note to the Trash — recoverable; emptying the Trash is manual in Zotero. Notes only, not items/collections/attachments. Requires a writable library (web API key or hybrid mode). Example: (action='create', item_key='ABC12345', note_title='Reading notes', note_text='Key claim ...').
Inputs: action (required), item_key (required), tags, append, +2 more. Returns: result
zotero_read_pdf_pagesA
Read specific page range(s) from a PDF attachment of a Zotero item. Use this when you know which pages to read — for example after getting the PDF outline via zotero_get_pdf_outline. Pages are 1-indexed. Returns Markdown with the page's heading structure preserved.
Inputs: item_key (required), start_page (required), end_page. Returns: result
zotero_search_by_citation_keyA
Look up a single Zotero item by its BetterBibTeX citation key (e.g. 'Smith2024' or 'cladderMicus2018'). Returns that one item's metadata, or a not-found message if no item has that key. citekey: the citation key exactly as assigned by BetterBibTeX (case-sensitive). In local mode: queries the running Better BibTeX plugin via its HTTP API (Zotero desktop must be running and have BBT installed). In web mode: scans the 'Extra' field of items for 'Citation Key:' lines — slower, and may miss items whose keys aren't persisted to Extra. Requires the Better BibTeX plugin in the user's Zotero install. For partial-key or free-text lookup, use zotero_search_items. Example: zotero_search_by_citation_key(citekey='hasan2026mcp') → metadata for that single item.
Inputs: citekey (required). Returns: result
zotero_search_by_tagA
Find items carrying one or more tags, with boolean syntax support. tag: list of tag strings; each entry is a condition ANDed with the others, and within an entry you can use ' OR ' for disjunction and a leading '-' for exclusion. Example: tag=['methods OR methodology', '-draft'] matches items tagged 'methods' OR 'methodology' AND NOT tagged 'draft'. item_type: '-attachment' (default) excludes attachments; pass 'journalArticle', 'book', etc. to filter. limit: max results (default 10). collection_key: optional 8-char key to scope to a collection. include_subcollections: also search collections nested beneath it (default False). Use zotero_get_tags to discover available tag names first. For free-text content search, use zotero_search_items or zotero_semantic_search instead. Example: zotero_search_by_tag(tag=['to-read'], limit=20).
Inputs: tag (required), limit, item_type, collection_key, +1 more. Returns: result
zotero_search_collectionsA
Search collections by name in the active library and return their 8-character keys. Matching is case-insensitive substring and applies ONLY to the collection's own name — not to parent names, descriptions, or items inside the collection. Multi-word queries are ANDed across words (NOT OR-ed): query 'reading list' matches only collections whose name contains both 'reading' AND 'list'. To match either word, issue two separate searches. Leading/trailing whitespace is ignored and empty words are dropped. Returns the collection's key plus its parent (if any). include_trashed: when True, also match collections currently in the Zotero Trash (results annotated as such). Default False — trashed collections are otherwise invisible to automated clients. Performance: scans all collections in the active library (O(n)); for very large libraries expect a full-list pagination under the hood. Example: zotero_search_collections(query="orals") → keys for every collection with "orals" in its name.
Inputs: query (required), include_trashed. Returns: result
zotero_search_itemsA
Search Zotero items by substring match against metadata (title, creators, year, and — in 'everything' mode — abstract). Returns metadata + abstracts as markdown. IMPORTANT: keep queries SHORT and SIMPLE — 'Author Year' (e.g. 'Brewer 2011') or just an author name ('Cladder-Micus'). This is substring matching, not web search: each extra word NARROWS the match, so adding topic words usually returns fewer results, not more. For topic discovery, use zotero_semantic_search instead; for tag filtering use zotero_search_by_tag. If a query finds nothing, this tool automatically falls back to simplified queries and then semantic search. query: required substring. qmode: 'titleCreatorYear' (default) matches only title/authors/year; 'everything' also searches abstract. item_type: '-attachment' (default) excludes attachments; pass 'journalArticle', 'book', etc. to filter. tag: optional list of tag conditions (ANDed). limit: max results (default 10). collection_key: 8-char key to restrict to a collection (bypasses the fallback cascade). include_subcollections: also search collections nested beneath it (default False). search_all_libraries: search personal + all group libraries at once, labelling each result with its library — use it when you don't know which library holds the item. Needs ZOTERO_SEARCH_BACKEND=sqlite; excludes collection_key. Example: zotero_search_items(query='Cladder-Micus') or zotero_search_items(query='Brewer 2011', search_all_libraries=True).
Inputs: query (required), tag, limit, qmode, +4 more. Returns: result
zotero_semantic_searchA
Prioritized topic-search tool. Find papers by semantic similarity to a query using AI embeddings — the BEST tool for finding papers on a topic (e.g. 'papers about mindfulness-based therapy'), far more efficient than scanning collection items or reading abstracts. Searches the ACTIVE library by default; pass search_all_libraries=True to cover every indexed library. query: the topic or concept; natural-language phrases work well. limit: max results (default 10). filters: optional metadata filters as a dict (e.g. {'itemType': 'journalArticle', 'year': '2023'}); also accepts a JSON string. library_id: optional — scope to one library other than the active one: 0 or 'user' for personal, else a groupID (see zotero_list_libraries). search_all_libraries: search every indexed library at once, labelling each result with its library; needs ZOTERO_SEARCH_BACKEND=sqlite, excludes library_id. Requires the semantic search database to be POPULATED — run zotero_update_search_database first if you just installed the server or added new items; check readiness with zotero_get_search_database_status. Available only when the [semantic] optional dependency is installed. Example: zotero_semantic_search(query='mindfulness-based cognitive therapy for depression', limit=5).
Inputs: query (required), limit, filters, library_id, +1 more. Returns: result
zotero_set_item_collectionsA
Change which collections existing items belong to — an incremental add/remove of item membership, NOT collection creation (use zotero_create_collection / zotero_delete_collection for that). item_keys must be an ARRAY of item keys, e.g. ["KEY1", "KEY2"] — not a single string. add_to and remove_from accept arrays of collection keys, names, or '/'-separated paths (resolved and validated automatically; unknown, trashed, or ambiguous specs fail before anything is changed). Existing memberships not named in remove_from are left alone; to replace an item's memberships wholesale use zotero_update_item. Use zotero_search_items to find item keys and zotero_search_collections to find collection keys.
Inputs: item_keys (required), add_to, remove_from. Returns: result
zotero_set_item_parentA
Set or clear the parent of a Zotero item. Pass a parent item key to assign or change the parent, or null to make the item top-level. Zotero validates whether the requested parent-child relationship is allowed.
Inputs: item_key (required), parent_key (required). Returns: result
zotero_switch_libraryA
Switch the active library context. EVERY subsequent read/write tool call (collections, items, annotations, search — all of them) operates on the library set here. Changes persist for the rest of the session or until the next switch. Discover valid library IDs/types via zotero_list_libraries first; don't guess. library_id: library ID string as returned by zotero_list_libraries (numeric for user/group, numeric for feeds). library_type: 'user' — the personal library; 'group' (default) — a group library; 'feeds' — a local RSS feed library; 'default' — RESET to whatever the ZOTERO_LIBRARY_ID / ZOTERO_LIBRARY_TYPE env vars configure (library_id is ignored in this mode). Fails fast if the library_id isn't accessible under the current credentials. Example: zotero_switch_library(library_id='5294983', library_type='group') or zotero_switch_library(library_id='', library_type='default').
Inputs: library_id (required), library_type. Returns: result
zotero_synthesize_annotationsA
Collect every highlight, annotation comment, and child note across a scope and organize them into a structured, per-paper digest that YOU (the agent) can then synthesize into a literature summary. This tool does NOT call an LLM — it only gathers and groups the raw material, so the synthesis step is yours. collection_key: optional 8-character collection key; when given, only annotations/notes whose resolved paper is a member of that collection are included. When omitted, the whole active library is scanned (capped by limit). tag: optional tag or list of tags to filter items by (accepts a string, a JSON list, or a list). limit: cap on annotations/notes scanned (default 200) to keep the call tractable. format='markdown' (default) groups the digest by paper; format='json' returns the same highlights and notes as structured records for downstream processing. Markdown output has each paper heading followed by its highlights (with attached comments) and any note excerpts — plus a top summary line counting papers, highlights, and notes. Use this before writing a thematic review so you can spot themes and contradictions across sources. Example: zotero_synthesize_annotations(collection_key='MT53KB66').
Inputs: tag, limit, format, collection_key. Returns: result
zotero_update_annotationA
Update an existing Zotero annotation. Editable fields: text (highlight text), comment, color (hex like '#ffd400'), and tags. Tags can be replaced wholesale via tags, or edited incrementally via add_tags/remove_tags (mutually exclusive with tags). Position/page/sortIndex are anchored to the PDF/EPUB geometry and are not editable.
Inputs: annotation_key (required), tags, text, color, +3 more. Returns: result
zotero_update_itemA
Update metadata on an existing Zotero item by key. Only what you pass is changed. fields: {name: value} of metadata to set (a JSON object string is accepted). Names may be snake_case (title, date, doi, url, abstract, publication_title, access_date, short_title, book_title, citation_key, item_type, place, extra, volume, issue, pages, publisher, issn, isbn, edition, language) or any raw Zotero API field name. An unknown name fails the call and lists the valid ones; a name that is not valid for this item's type is reported as skipped. item_type migrates the item (overlapping fields kept, type-specific ones dropped). TAG SEMANTICS (easy to get wrong): tags REPLACES the whole tag list; add_tags/remove_tags are incremental and preferred. They are mutually exclusive with tags. collections (keys) and collection_names likewise REPLACE membership — pass collections=[] to clear it; for incremental moves use zotero_set_item_collections. creators: full replacement list of {creatorType, firstName, lastName} objects. Requires a writable library (fails in local-only mode). To edit notes use zotero_manage_note. Example: zotero_update_item(item_key='RTKZQI8E', fields={'doi': '10.1145/3708319'}, add_tags=['reviewed']).
Inputs: item_key (required), tags, fields, add_tags, +4 more. Returns: result
zotero_update_search_databaseA
Build or refresh the semantic search embedding database from Zotero items. Run this: (a) after first install, (b) after adding items via zotero_add_item, or (c) when the user has added items directly in Zotero desktop since the last update. By default the update is INCREMENTAL — only new or changed items are re-embedded, so repeated calls are cheap. force_rebuild=True re-embeds ALL items from scratch (slow; use when changing the embedding model or recovering from corruption). limit: optional cap on items processed (useful for smoke-testing). Progress is reported via the MCP context; on large libraries an incremental update is seconds, a full rebuild can take minutes. Requires the [semantic] optional dependency and a configured embedding provider (see config.json). Check status with zotero_get_search_database_status. Example: zotero_update_search_database() after adding a batch of papers.
Inputs: limit, force_rebuild. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 38 tools
The tools are largely distinct resource/action pairs, and descriptions repeatedly steer agents toward the correct search tool (semantic vs substring vs tag vs advanced). A few close pairs remain, such as zotero_get_annotations vs zotero_synthesize_annotations and zotero_add_item vs zotero_attach_file, but their intended uses are clearly separated.
All 38 tools follow a uniform zotero_<verb>_<object> snake_case pattern with consistent verbs like get, create, update, delete, search, and set. The few generic names, such as zotero_manage_note and zotero_get_recent, do not break the overall pattern.
38 tools is large, but the scope matches Zotero's breadth: items, collections, annotations, notes, search, PDFs, citations, libraries, and semantic search. It is slightly heavier than ideal and has some redundancy among search and annotation readers, but each tool addresses a distinct workflow.
Core CRUD is covered for items, notes, annotations, and collections, along with search, file attachment, PDF reading, bibliography export, and library switching. Minor gaps remain: collections cannot be renamed/updated, there is no restore-from-trash tool, and zotero_delete_item references a zotero_delete_note tool that does not exist, though zotero_manage_note covers note deletion.
Maintenance
Related MCP Connectors
Zotero MCP server for Claude and ChatGPT: search, citations, safe writes, PDF passages and pages.
Academic literature search, retrieval, and private library management on top of OpenAlex.
Personal context for every AI: search, read, and write back to your private Markdown library.
Your personal library, readable by AI agents: durable snapshots, hybrid search, 46 MCP tools.
Related MCP Servers
- AlicenseAqualityCmaintenanceEnables AI assistants to read, search, and fully manage a Zotero library locally—organizing collections, editing items, uploading files, and managing notes, tags, and trash—through the local Zotero API without cloud sync.331MIT
- FlicenseNot gradedqualityAmaintenanceEnables AI assistants to search a Zotero library and retrieve item metadata or full-text PDF contents via the Model Context Protocol.-
- FlicenseNot gradedqualityAmaintenanceEnables AI assistants to search a Zotero library, retrieve complete item metadata, and access full text from stored PDFs through the Model Context Protocol.-
- FlicenseNot gradedqualityAmaintenanceEnables AI assistants to search a Zotero library, retrieve complete item metadata, and fetch full text from attached documents such as PDFs through the Model Context Protocol.-