The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Sothebys listing page.
Realized auction prices from Sotheby's — search the past-lot archive by artist or maker across every sale, and get the price including buyer's premium, the winning (hammer) bid, the estimate range and the sale's name/date/location; pull full catalogue detail (description, provenance, literature, exhibition history, images) for one lot.
Part of Pipeworx — an MCP gateway connecting AI agents to 1683+ live data sources.
sothebys_results_search(query, category?, sale_type?, page?, limit?, include_unpriced?)
— search the archive by artist/maker/keyword. category is a free-text
filter matched against the lot's department names (e.g. "prints",
"photographs"). sale_type is "timed" (default), "live" or "any" —
see What is and isn't public below, it is the single most important
argument here.sothebys_lot_details(lot_id) — full catalogue detail for one lot, keyed by
the lot_id a sothebys_results_search result carries.Tools are prefixed with the house name deliberately: three art-house packs
(Bonhams, Sotheby's, Phillips) expose the same two shapes, so bare
results_search would collide across them and make ask_pipeworx routing
ambiguous.
Keyless. Both upstreams serve anonymous callers; the Algolia search key is minted on demand by Sotheby's own public GraphQL and cached for 5 minutes.
Sotheby's is not the Bonhams shape. Its lot pages are a Next.js app fed by
a separate GraphQL backend, so there is no server-rendered search page to read.
Two public, keyless endpoints carry everything, and neither is /bsp-api/*:
https://clientapi.prod.sothelabs.com/graphql — Sotheby's own public bidding
API. Unauthenticated POST works. Introspection is disabled by its Cosmo
Router, but the operation documents ship in the site's own JS bundles under
/buy/_next/static/chunks/. Three operations are used:
AlgoliaSearchKeyQuery → a search key for the index below. Requesting no
filters yields a key scoped to the whole index, not one auction.LotCardsQuery(id, lotIds, language) → auction.lotCardsFromIds, the
per-auction batch price lookup. bidState.currentBidV2 is the winning
(hammer) bid; bidState.sold.premiums.finalPriceV2 is the realized price
including buyer's premium. Amounts arrive as decimal strings.
auction.testRecord flags Sotheby's QA sales, which are in the index.lotV2(lotId, language, countryOfOrigin) → full catalogue detail. It
returns the Lottable interface, so the selection needs
... on LotV2 { … }; condition is an object, not a string.https://KAR1UEUPJD-dsn.algolia.net/1/indexes/prod_lots/query — the search
index the site itself queries (window.ALGOLIA_APP_ID /
window.ALGOLIA_INDEX_NAME in the page source). This is the only
cross-auction search path. It carries creators, departments, estimates,
auctionType, auctionDate and the site-relative lot slug — but no
prices at all: price is null on every record, which is why every search
costs a second, per-auction GraphQL round trip.Neither host is www.sothebys.com, so the Crawl-delay: 15 and the
Disallow: /bsp-api/* in https://www.sothebys.com/robots.txt (checked
2026-09-05) do not cover them; clientapi.prod.sothelabs.com serves no
robots.txt at all (404). This pack makes no request to www.sothebys.com —
it only builds www URLs for a human to click.
The realized price comes from the GraphQL bidState.sold union, which is
either ResultVisible or ResultHidden depending on whether Sotheby's
publishes that sale's results through this platform. Measured 2026-09-05
across the 44 distinct sales in a 100-hit "Pablo Picasso" sample, the split is
a clean rule rather than noise:
| slice | sales visible |
|---|---|
auctionType: Live (saleroom sales), any year including 2025–26 | 0 of 11 |
auctionType: Timed (online sales), ~2021 onwards | 18 of 18, every lot |
auctionType: Timed, before ~2021 | 0 of 15 |
So sale_type defaults to timed: that is the slice of the archive where a
realized price exists at all. Asking for live returns lots with null prices
however famous they are — Sotheby's publishes live-saleroom results on its own
results pages, not through this API — and the tool says so in note rather
than letting it read as a parse failure.
Two consequences worth knowing when reading a response:
sothebys_results_search scans a page of up to 100 search hits and returns
the priced subset, so count is normally well below scanned, and
total_hits (the whole archive match count) is far larger than both.bonhams — same two-tool shape for Bonhams, whose realized prices are plain
server-rendered HTML.Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):
tools/list at https://gateway.pipeworx.io/sothebys/mcp returns the tools in the table
above plus the shared Pipeworx meta-tools — ask_pipeworx,
discover_tools, search_within, remember/recall and the rest of the
gateway-wide set. So the tool count you see is larger than this table: a
single-pack endpoint currently lists roughly 30 shared tools alongside the
pack's own. The connection's initialize response states its exact scope, and
is the authoritative answer for a given day.
This is deliberate, not multiplexing by accident. The meta-tools are what let a
scoped connection answer a question this pack does not cover — via
ask_pipeworx, which routes across the whole catalog — without you adding a
second MCP server. There is currently no way to mount a pack endpoint without
them; if the extra schemas cost you more context than the routing is worth,
connect to the full gateway once rather than to several pack endpoints.
Or connect to the full Pipeworx gateway to get every pack's tools listed directly, instead of just this one's:
Both URLs reach the same gateway and the same 1683+ data sources. The
only difference is which pack's tools are listed directly; ask_pipeworx
reaches all of them from either one.
No account needed for the first calls. Inspect any tool: GET https://gateway.pipeworx.io/v1/tools/sothebys_results_search. Find one: POST https://gateway.pipeworx.io/v1/tools/search_packs with {"query":"..."}.
This package also runs as a local stdio MCP server — no Pipeworx account, no gateway round-trip:
Or run it directly to confirm it starts:
It speaks MCP over stdin/stdout and answers initialize/tools/list/tools/call
for only this pack's tools — none of the shared meta-tools the gateway
connection above adds. Same source, same tools, no ask_pipeworx routing.
Instead of calling tools directly, you can ask questions in plain English — this works on the pack endpoint above as well as on the full gateway:
The gateway picks the right tool and fills the arguments automatically.
MIT