Library, archive, and museum MCP servers let AI assistants manage research citations, search ebook collections, retrieve archived web pages, and explore museum art. Instead of manually exporting Zotero bibliographies, navigating Calibre’s interface, or browsing museum websites one artwork at a time, these servers let AI agents query your research library, extract ebook chapters, access historical snapshots of websites, and discover artworks across major museum collections — all through the Model Context Protocol.

This review covers the library, archive, and museum vertical — reference management (Zotero), ebook management (Calibre, EPUB/PDF), book discovery (Open Library, Google Books), web archiving (Wayback Machine), digital archive standards (IIIF), and museum collections (Rijksmuseum, Met, Smithsonian, Art Institute of Chicago, Harvard Art Museums, Cooper Hewitt). For academic paper search (arXiv, Semantic Scholar, PubMed), see our Education & EdTech MCP review. For government document access (Congress.gov, GovInfo), see our Government & Public Sector MCP review.

The headline findings: Zotero’s growth is accelerating — 54yyyu/zotero-mcp jumped from 1,800 to 2,700 stars in six weeks, with 10+ implementations now available. NEW: ZotLink (134 stars) adds production-ready preprint saving from arXiv, bioRxiv, medRxiv, chemRxiv, and CVF. NEW: IIIF MCP server provides 20+ tools for the International Image Interoperability Framework — the universal standard used by thousands of cultural heritage institutions. Six major museums now have community MCP servers (up from five) with the addition of Cooper Hewitt. Traditional library systems (ILS, OPAC, MARC) still have zero MCP presence — a significant gap given how many libraries exist worldwide.

Category: Education & Learning

Reference Management (Zotero)

54yyyu/zotero-mcp

Server Stars Language License Tools
54yyyu/zotero-mcp 2,700 Python MIT 13+

The most popular library MCP server by far and one of the highest-starred MCP servers in any niche vertical. Connects your Zotero research library with AI assistants via the Zotero local API. Key capabilities:

  • Semantic search — vector-based similarity search across your entire library with multiple embedding models (default free model, OpenAI, or Gemini)
  • Auto-updating database — configurable sync schedules keep the search index current
  • Advanced search — filter by collections, tags, authors, date ranges
  • Content extraction — full-text retrieval, metadata access, and PDF annotation extraction including image annotations
  • Collection browsing — navigate your library’s organizational structure

The semantic search is the killer feature — instead of keyword matching, you can ask conceptual questions like “papers about transformer attention mechanisms” and get relevant results ranked by embedding similarity. The auto-sync means you don’t need to manually rebuild the index after adding new papers.

cookjohn/zotero-mcp

Server Stars Language License Tools
cookjohn/zotero-mcp 684 TypeScript MIT 20

Takes a fundamentally different architecture — runs as a native Zotero plugin rather than an external server. The plugin embeds an MCP server using Streamable HTTP protocol, so AI clients connect directly to Zotero. The 20 tools span four categories:

  • Search & Query (7 tools) — library search, annotation search, full-text search, collection browsing, item details, abstracts, and content extraction with four detail levels (minimal/preview/standard/complete)
  • Collection Management (4 tools) — collections, subcollections, collection items, and details
  • Semantic Search (3 tools) — vector search with OpenAI or Ollama support, similarity matching, and index status
  • Write Operations (4 tools) — create notes, add tags, update metadata, and write new items

The write capability is the key differentiator — 54yyyu’s server is read-only, while cookjohn’s can create notes, tag items, and update metadata directly in your Zotero library. The annotation system supports color and tag filtering. Includes a built-in client configuration generator.

TonybotNi/ZotLink — Preprint Saving to Zotero

Server Stars Language License Tools
TonybotNi/ZotLink 134 Python MIT 5

A production-ready MCP server for saving open preprints to Zotero with rich metadata and smart PDF attachments. Covers arXiv, CVF OpenAccess, bioRxiv, medRxiv, and chemRxiv — with publisher databases (Nature, Science, IEEE Xplore, Springer) planned. The 5 tools:

  • check_zotero_status — verify Zotero connection
  • get_zotero_collections — browse library structure
  • save_paper_to_zotero — save with full metadata extraction
  • extract_arxiv_metadata — pull metadata from arXiv papers

At 134 stars, ZotLink is the third most popular Zotero MCP server. Its focus on preprint ingestion complements the broader library access of 54yyyu and cookjohn’s implementations — where those servers help you read your library, ZotLink helps you build it.

Additional Zotero Implementations

More Zotero MCP servers exist, each with a slightly different focus:

With 10+ independent implementations, Zotero is one of the most MCP-served single applications in the entire ecosystem — rivaling even Slack and GitHub for community attention. The 54yyyu server’s jump from 1,800 to 2,700 stars in six weeks shows research-AI integration demand is accelerating.

eBook Management

onebirdrocks/ebook-mcp

Server Stars Language License Tools
onebirdrocks/ebook-mcp 361 Python Apache 2.0 10

The most popular ebook MCP server — works directly with EPUB and PDF files without requiring any library management software. The 10 tools split across two format categories:

EPUB tools:

  • get_all_epub_files — locate EPUB documents in specified directories
  • get_metadata — extract publication info (title, author, publisher, language, ISBN)
  • get_toc — retrieve table of contents structure
  • get_chapter_markdown — convert chapter content to Markdown

PDF tools:

  • get_all_pdf_files — discover PDF documents
  • get_pdf_metadata — access document properties
  • get_pdf_toc — extract outline and chapter structure
  • get_pdf_page_text — retrieve plain text from specific pages
  • get_pdf_page_markdown — convert pages to Markdown
  • get_pdf_chapter_content — extract chapters by title with page references

The chapter-level extraction is the standout capability — instead of dumping an entire book into context, you can selectively pull specific chapters or pages. Uses ebooklib for EPUB, PyPDF2 and PyMuPDF for PDF, with BeautifulSoup and html2text for content conversion.

trieloff/calibre-mcp

Server Stars Language License Tools
trieloff/calibre-mcp 28 Shell Apache 2.0 4

A pure bash implementation that uses the calibredb command-line interface — no Python or Node.js dependencies beyond Calibre itself. The four tools cover the essentials:

  • Full-text search with automatic phrase matching and fuzzy fallback
  • Author search with partial matching
  • Title search with partial matching
  • Excerpt extraction with keyword highlighting and pagination

Provides both Calibre deep links (calibre://) and file URLs (file://) for results. Includes timeout handling optimized for macOS. The bash-only approach makes it the simplest Calibre MCP server to install and understand.

Additional Calibre Implementations

  • sandraschi/calibremcp (Python, MIT, 21 tools) — the most feature-rich Calibre server with FastMCP, including RAG retrieval, LanceDB metadata indexing, natural language search, and agentic workflows
  • ajtudela/calibre_mcp_server (Python) — search and retrieve book metadata through MCP
  • ispyridis/calibre-mcp-nodejs (Node.js) — supports EPUB, PDF, MOBI, AZW/AZW3, TXT, HTML and more formats
  • THeK3nger/calibre-mcp — queries Calibre via the Content Server, with full-text search and metadata retrieval
  • pshap/mcp-neolibrarian (Python, MIT, 14 tools) — read-only server with unified search, batch operations, content analysis, and FTS statistics
  • benoute/calibre-mcp (Python) — access to a local Calibre library with search and retrieval

Book Discovery

8enSmith/mcp-open-library

Server Stars Language License Tools
8enSmith/mcp-open-library 70 TypeScript MIT 6

Connects to the Internet Archive’s Open Library API — the largest open book database with records for over 20 million editions. The 6 tools cover:

  • Book search by title — find books with metadata
  • Author search by name — locate authors
  • Author info — detailed author data via Open Library key
  • Author photo — photo URLs via OLID
  • Book cover — cover image URLs from ISBN, LCCN, OCLC, or OLID
  • Book by ID — comprehensive details from multiple identifier types

No API key required — Open Library’s API is free and open. A solid choice for book metadata enrichment, reading list creation, or bibliographic research.

Additional Book Discovery Servers

  • juanbeniteza/mcp-google-books (Python, MIT, 2 tools) — book and author search via Google Books API
  • andylbrummer/booklife-mcp (Go, 27 tools) — unifies Hardcover (reading tracker), Libby/OverDrive (library borrowing), and Open Library (metadata) into one reading assistant with a unified TBR list, reading history, and recommendations
  • CodeDreamer06/BookDownloaderMCP (TypeScript) — search and download ebooks from LibGen mirrors with smart mirror discovery and failover

Internet Archive & Wayback Machine

Mearman/mcp-wayback-machine

Server Stars Language License Tools
Mearman/mcp-wayback-machine 21 TypeScript CC BY-NC-SA 4.0 4

An MCP server and CLI tool for the Wayback Machine — no API keys required. The 4 tools:

  • save_url — archive a URL to the Wayback Machine, returns archived URL with timestamp
  • get_archived_url — retrieve an archived version by URL and optional timestamp
  • search_archives — query all available snapshots with date range filtering
  • check_archive_status — archival statistics including first/last capture, total captures, and yearly breakdown

Built-in rate limiting at 15 requests per minute protects against API abuse. Works as both an MCP server for AI clients and a standalone CLI tool. The CC BY-NC-SA license is worth noting — it restricts commercial use, which may matter for some deployments.

sisilet/wayback-mcp

Server Stars Language License Tools
sisilet/wayback-mcp 2 Python MIT 3

A simpler alternative with MIT licensing. Uses the CDX API for snapshot retrieval and adds archive.org item search beyond just the Wayback Machine. Provides a wayback://{url}/{timestamp} resource protocol for direct access to archived content.

Digital Archive Standards (IIIF)

code4history/IIIF_MCP — Universal Cultural Heritage Access

Server Stars Language License Tools
code4history/IIIF_MCP 1 JavaScript — 20+

The first MCP bridge to the IIIF ecosystem — and potentially the most impactful server in this entire category. IIIF (International Image Interoperability Framework) is the universal standard used by thousands of cultural institutions worldwide — the Library of Congress, British Library, Bibliothèque nationale de France, Harvard, Stanford, the Vatican Library, and thousands more publish their digitized collections through IIIF APIs. This single MCP server can access them all.

The 20+ tools span the full IIIF specification:

  • Content Search — full-text search within IIIF documents (manuscripts, books, newspapers)
  • Metadata Retrieval — structured metadata from IIIF manifests and collections
  • Image Operations — region, size, rotation, and quality parameters per the IIIF Image API
  • Collection Navigation — browse hierarchical collection structures
  • Annotation Management — read and manage annotations on digitized content
  • Activity Streams — track changes and updates across collections
  • Authentication — handle IIIF Auth flows for restricted content
  • Audio/Video — support for time-based media resources (IIIF Presentation API 3.0)

The standout capability is universality — rather than building a separate MCP server for each museum or library (as the museum servers below do), IIIF_MCP works with any IIIF-compliant institution. The v1.1.0 release added image data fetching and canvas operations. Available as a single-file bundle for easy deployment.

The star count is low (1 star) but this is a genuinely important bridge — IIIF covers more cultural heritage content than all the museum-specific MCP servers combined.

Museum & Cultural Heritage Collections

r-huijts/rijksmuseum-mcp

Server Stars Language License Tools
r-huijts/rijksmuseum-mcp 67 JavaScript/TypeScript MIT 7

The most popular museum MCP server. Provides access to the Rijksmuseum’s collection through 7 tools:

  • search_artwork — find artworks by text, artist, type, materials, periods, and colors
  • get_artwork_details — comprehensive info including provenance and exhibition history
  • get_artwork_image — high-resolution images with tile-based loading and multiple zoom levels
  • get_user_sets — browse user-created collections and thematic groupings
  • get_user_set_details — detailed info about curated sets
  • open_image_in_browser — display artwork images for detailed viewing
  • get_artist_timeline — chronological timeline of an artist’s works tracking style evolution

The artist timeline and tile-based image loading are unique features not found in other museum MCP servers. Requires a Rijksmuseum API key (free to obtain).

mikechao/metmuseum-mcp

Server Stars Language License Tools
mikechao/metmuseum-mcp 26 TypeScript MIT 4

Access to The Metropolitan Museum of Art’s collection with an interactive MCP App — a browsable art explorer that runs directly within MCP-compatible clients. The 4 tools:

  • list-departments — all departments at The Met
  • search-museum-objects — collection search with filters for images, departments, and pagination
  • get-museum-object — detailed artwork info with optional base64-encoded images
  • open-met-explorer — launches the interactive browsing application

The MCP App (ui://met/explorer.html) for in-chat browsing is the standout feature. Uses The Met’s free public API — no key required.

molanojustin/smithsonian-mcp

Server Stars Language License Tools
molanojustin/smithsonian-mcp 8 Python — 16

The most comprehensive museum MCP server with 16 tools spanning all 21 Smithsonian museums and their combined 155+ million objects. Organized into two categories:

Search & Discovery (11 tools) — smart sampling across museums, advanced filtering, museum-specific queries, on-view status checking, and object type browsing.

Information & Context (5 tools) — museum directory, collection statistics with per-museum breakdowns, and context retrieval for search results and objects.

The get_objects_on_view tool is unique — it can tell you what’s currently exhibited in any Smithsonian museum, useful for trip planning or identifying objects you can actually go see.

Additional Museum Servers

  • mikechao/artic-mcp (TypeScript, MIT, 7 tools) — Art Institute of Chicago collection with full-text search, artist browsing, medium search, and an art gallery prompt that generates interactive HTML galleries
  • AlexLin1234/harvard-art-museums-mcp-server (Python, MIT, 5 tools) — access to 224,000+ artworks, artists, and museum objects with random artwork discovery
  • behole/cooper-hewitt-mcp (TypeScript, MIT, 2 tools) — NEW. Cooper Hewitt, Smithsonian Design Museum collection access with object search and detailed retrieval. The first design-focused museum MCP server — Cooper Hewitt’s collection emphasizes design objects, textiles, wallcoverings, drawings, and product design rather than fine art

Six world-class museums now have community-built MCP servers — up from five. None has built an official one, but the community coverage is solid — these are all free public APIs wrapped in clean MCP interfaces. Combined with the IIIF MCP server above, AI agents can now access both museum-specific collections and the broader universe of IIIF-compliant digital archives.

What’s missing

The gaps in this category are still significant, though narrowing:

  • No Library of Congress digital collections MCP — congress.gov API covers legislative data only, not the LC’s vast digital collections (prints, photographs, maps, manuscripts, recordings). The IIIF MCP server can theoretically access LC’s IIIF endpoints, but no dedicated server exists
  • No Europeana or DPLA servers — the two largest digital cultural heritage aggregators (60+ million items combined) have public APIs but no dedicated MCP servers. Again, IIIF_MCP can access Europeana’s IIIF resources, but aggregator-level search is missing
  • No MARC, Dublin Core, or Z39.50 servers — zero MCP presence for bibliographic cataloging standards
  • No ILS integrations — Koha, FOLIO, Alma, Sierra, and other library management systems have no MCP servers
  • No digital preservation servers — no integration with preservation platforms like Archivematica, DSpace, or Fedora
  • No public library catalog access — no way for an AI agent to search a local public library’s collection via MCP

The IIIF MCP server is a significant step toward bridging the institutional gap — it provides a universal protocol layer that works with thousands of institutions. But it’s a low-level tool (you need to know IIIF manifest URLs) rather than a discovery layer. The absence of institutional library infrastructure MCP servers remains the category’s biggest gap.

Bottom line

The library, archive, and museum MCP category earns 3.5/5. It’s deep where individuals need it — Zotero’s 10+ implementations and 2,700-star lead server (up 50% in six weeks) rival the best-served applications in the entire MCP ecosystem, and Calibre’s 6+ servers cover ebook management thoroughly. Museum collections now span 6 major institutions with the addition of Cooper Hewitt. The IIIF MCP server is a significant new addition — providing universal access to thousands of cultural institutions through a single protocol. But the category still lacks institutional library infrastructure — no ILS, no cataloging standards, no public library access, no dedicated digital heritage aggregator servers. If you’re a researcher with a Zotero library or a reader with a Calibre collection, you’re well-served. If you work in library science or cultural heritage at an institutional level, IIIF_MCP is a promising start but MCP hasn’t fully arrived yet.

This review was last refreshed on 2026-04-26 using Claude Opus 4.6 (Anthropic).