Library and Museum Agent Readiness: Why Cultural Institutions Are Missing the AI Discovery Wave
There are over 55,000 libraries and museums in the United States. They hold some of the richest structured data on the planet — catalog records, collection metadata, event schedules, educational programs. Almost none of it is accessible to AI agents. When a travel agent asks “what exhibits are open near me this weekend,” it gets silence.
The Paradox: Rich Data, Zero Access
Libraries and museums are paradoxically some of the most data-rich and least agent-accessible institutions in the economy. A university library catalogs millions of items with standardized metadata — MARC records, Dublin Core, controlled vocabularies. A natural history museum tags every specimen with taxonomy, provenance, collection date, and conservation status. This is structured data that would make most SaaS companies envious.
But the data sits behind human-only interfaces. The OPAC (Online Public Access Catalog) is designed for a person typing keywords into a search box. The museum collection database powers a web gallery that requires clicking and scrolling. Event calendars are HTML pages. Program registration is a phone call or an email to the education department.
For AI agents, this means cultural institutions are effectively dark. An AI travel planner cannot check which exhibits are currently showing at a local museum. An AI research assistant cannot search a library's special collections programmatically. An AI education agent cannot enroll a child in a summer reading program. The data exists. The access does not.
What Agent-Ready Looks Like for Cultural Institutions
Four capabilities separate an agent-invisible institution from one that AI agents can discover, search, and transact with.
Collection Search
Current State
OPAC web interface requiring human interaction, keyword search behind login walls
Agent-Ready
Public collection search API with filters for medium, period, artist, availability
Event Calendar
Current State
HTML event pages, PDF flyers, social media posts for upcoming programs
Agent-Ready
Structured event endpoint with dates, descriptions, capacity, registration status
Ticket / Membership
Current State
Ticket purchase through third-party widget or phone, membership forms as PDFs
Agent-Ready
Ticket availability API with pricing tiers, membership enrollment endpoint with plan comparison
Educational Programs
Current State
Program listings on web pages, registration by email or phone, waitlists managed manually
Agent-Ready
Program catalog API with age groups, topics, capacity, registration endpoint with confirmation
The Exceptions: Who Is Doing It Right
A handful of major institutions have public APIs that make portions of their collections searchable by machines. They prove the model works — but they also highlight how far the sector has to go.
Library of Congress
Est. 52Public REST API for digital collections (loc.gov/apis). JSON responses, search by subject, date, format. No ticket booking needed. Missing: agent-card.json, MCP server, llms.txt.
Smithsonian Open Access
Est. 48Open Access API covers 4.5M+ objects across 21 museums. CC0 licensing. JSON with rich metadata. Missing: event/ticket endpoints, agent-native discovery files, MCP.
Europeana
Est. 45REST API across 3,000+ institutions. 58M+ cultural objects searchable via API. Good documentation. Missing: transactional endpoints, agent-card, MCP.
Average Local Library
Est. 4Website with hours, location, maybe an OPAC link. No public API. No structured data beyond basic Schema.org. Phone number for everything else.
The gap between leaders and the field: Even the best cultural institution APIs (Library of Congress, Smithsonian) would score Silver at best on the Agent Readiness Score. They have excellent read-only collection data but lack agent-native discovery files, transactional capabilities, and MCP servers. The average local library or museum scores under 5 — website, phone number, and nothing else.
Why This Matters Now: The AI Cultural Agent
AI agents are already being built for travel planning, education, and research — three domains where libraries and museums are central. Consider the use cases that are emerging right now:
AI Travel Planners
A family visiting a new city asks their AI agent for museum recommendations. The agent needs exhibit data, hours, ticket prices, and booking capability. Without an API, it can only link to the website and say "check their hours."
AI Research Assistants
A historian asks their AI to find primary source documents about the Civil War. The Library of Congress API can serve this. The other 16,999 public libraries cannot. Millions of unique special collections are invisible.
AI Education Agents
A parent asks their AI to find age-appropriate weekend programs for their 8-year-old. Libraries and museums run thousands of programs. None are discoverable via API. The agent defaults to commercial alternatives.
AI Accessibility Agents
An agent helping a wheelchair user plan a museum visit needs to check accessibility features, elevator locations, and accessible entrance availability. This data exists — in PDFs and staff knowledge, not in APIs.
The risk for cultural institutions is not that AI agents will replace them — it is that agents will route people elsewhere. When an AI agent cannot find structured data about a museum exhibit, it recommends the commercial alternative that does have an API. The museum does not lose to another museum. It loses to a theme park or an online experience that is agent-accessible.
This is the same dynamic we documented in the education sector analysis — institutions with deep knowledge but no agent-facing infrastructure are ceding ground to commercial platforms that prioritize machine accessibility.
The Path Forward: From Dark to Discoverable
Cultural institutions do not need to build everything at once. The path from ARL-0 (Dark) to ARL-3 (Integrated) follows a practical sequence.
Discovery files (1 day)
Create agent-card.json and llms.txt describing the institution, collections, hours, and services. This alone moves from ARL-0 to ARL-1.
Event calendar API (1 week)
Expose upcoming events, exhibits, and programs as a structured JSON endpoint. Include dates, descriptions, capacity, and registration status.
Collection search endpoint (2 weeks)
If the OPAC or collection database exists internally, add a public REST endpoint with search, filtering, and pagination. Most catalog systems support this with configuration.
Transactional capabilities (1 month)
Ticket availability and booking, program registration, membership enrollment. These require integration with existing ticketing and registration systems.
MCP server (built on top of the above)
Wrap all endpoints in an MCP server with tool descriptions agents can discover automatically. AgentHermes can auto-generate this from existing APIs.
The institutions that move first will capture a disproportionate share of agent-driven cultural recommendations. There are parallels in the government sector — public institutions with valuable data that remains locked behind legacy interfaces.
Frequently Asked Questions
Why would an AI agent need to interact with a library or museum?
AI travel agents recommend cultural attractions based on user interests. AI research agents search collections for specific artifacts or publications. AI education agents find programs for children and adults. AI personal assistants book tickets, check hours, and reserve event spots. Every interaction that currently requires a phone call or website visit is a candidate for agent automation.
Do libraries have APIs?
A handful of major institutions do. The Library of Congress has a public REST API. Many academic libraries expose their catalogs through Z39.50 or SRU protocols, which are machine-readable but not agent-friendly (they predate modern REST conventions). The vast majority of public and local libraries have no API at all — just a website and an OPAC interface designed for human browsers.
What is the fastest way for a museum to become agent-ready?
Start with discovery files: agent-card.json and llms.txt describing your institution, collections, and services. Then expose your event calendar as a structured JSON endpoint. If you sell tickets, add an availability and booking API. These three steps can move a museum from ARL-0 (Dark) to ARL-2 (Basic) in a matter of weeks.
Is the Smithsonian agent-ready?
Partially. The Smithsonian Open Access API is excellent for collection search — 4.5 million objects with rich metadata and CC0 licensing. But it lacks transactional capabilities (no ticket booking, no event registration) and agent-native discovery files (no agent-card.json, no MCP server, no llms.txt). We estimate it at roughly 48 on the Agent Readiness Score — strong on data, weak on agent-native protocols.
How many libraries and museums are there in the US?
There are approximately 17,000 public libraries, 3,300 academic libraries, and 35,000 museums in the United States. That is over 55,000 cultural institutions, nearly all of which have zero agent-facing infrastructure. The cultural sector represents one of the largest untapped verticals in the agent economy.
Is your institution visible to AI agents?
Run a free Agent Readiness Scan to see how your library or museum scores across all 9 dimensions. Most cultural institutions score under 10.