Met Museum Scraper

Searches the Metropolitan Museum of Art's collection by keyword through its public Collection API. No login, no proxy, no browser.

Input

Field	Type	Notes
query	string	Required. Free-text search, e.g. sunflowers. Default sunflowers.
maxItems	integer	Cap on artworks returned. Default 10.
hasImages	boolean	Restrict to objects that have an image. Default true.

Output

One row per artwork: objectId, title, artist, artistNationality, objectDate, medium, dimensions, department, classification, culture, isPublicDomain, primaryImage, creditLine, accessionNumber, objectUrl.

Notes

• Search lives on v1.1; object detail still lives on v1. The Met retired the Solr-backed /v1/search on 2026-10-01 — it answers HTTP 410 with a pointer to the Elastic-backed /v1.1/search, which is paginated by offset and limit and caps a page at 500. There is no /v1.1/objects, so the two halves of a run deliberately use different API versions.
• An empty page comes back as objectIDs: null, not []. Both a query that matches nothing and an offset past the end return null, so the result is normalised before anything iterates it.
• The Met splits this into two endpoints: search returns only object IDs, and each object costs its own request. maxItems is therefore the main lever on how long a run takes.
• The object endpoint is behind a WAF that rate limits bursts. Firing requests back to back returned HTTP 403 with an HTML challenge page for 39 of 100 objects, and the block then applied to every request from that IP for about 30 seconds. Requests are now spaced 250 ms apart and retried with a capped backoff, which takes the same run to 100 of 100. Adding a User-Agent was tested and made no difference — this is purely about rate.
• Because a block is served as HTML rather than JSON, responses are parsed defensively; an HTML body is treated as a retryable block rather than crashing the run.
• The API returns "" rather than null for fields an object has no value for; those are normalized to null so an empty field is not mistaken for a real value.
• A query matching nothing throws rather than finishing with an empty dataset, which would otherwise look like a successful run that found nothing.