mcpField
B
Bonhams

Bonhams

Remote
pipeworx.ioBusiness & Marketing0Updated 5d ago

About Bonhams

Bonhams is an MCP server for AI agents and researchers to search Bonhams’ past auction lots and inspect individual lots, including prices and sale details.

Key features

  1. 01Search sold lots by artist or keyword
  2. 02Filter results by department
  3. 03Retrieve hammer and realized prices
  4. 04View estimates, date, and department
  5. 05Fetch lot descriptions, condition, and images

Use cases

  1. 01Research an artist’s past auction results
  2. 02Compare realized prices with estimates
  3. 03Inspect a specific lot’s description and images

MCP support

Exposes search and lot-detail tools to MCP clients. It can run as a local stdio MCP server or be accessed through the Pipeworx gateway.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://gateway.pipeworx.io/bonhams/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

Details

From the project README

@pipeworx/bonhams

Realized auction prices from Bonhams — search past ("sold") lots by artist or maker and get the hammer price, the hammer price including buyer's premium (the actual realized/sold price), estimate range, sale date and department; pull full lot detail (description, condition, images) for one lot.

Part of Pipeworx — an MCP gateway connecting AI agents to 1875+ live data sources. This is an independent, unofficial integration — not affiliated with, endorsed by, or published by the upstream provider.

Tools

  • bonhams_results_search(query, category?, page?, limit?) — search Bonhams' sold-lot archive by artist/maker/keyword. category is a free-text filter matched against the department name (e.g. "jewellery", "prints", "books").
  • bonhams_lot_details(auction_id, lot_no, slug?) — full detail for one lot, keyed by the auction_id/lot_no a bonhams_results_search result carries.

Auth

Keyless. Public pages, no login required.

Data sources

  • https://www.bonhams.com/search/?chronology=past&query=<q>&page=<n> — the site's own past-lots search page. It's a Next.js app that server-renders the FULL result set into a __NEXT_DATA__ JSON blob (pageProps.lotData.auctionLots[]) — a plain fetch() gets real data, no separate XHR/API call needed. Each entry carries price.hammerPrice (before premium) and price.hammerPremium (the realized price a buyer actually paid, matching the "Sold for $X inc. premium" text shown on the page) plus estimateLow/estimateHigh, department.name, hammerTime.datetime, auctionId, and lotNo (which is {full, letter, number} — full is sometimes absent, fall back to number).
  • https://www.bonhams.com/auction/<auctionId>/lot/<lotNo>/<slug>/ — the lot detail page, same SSR pattern, one "lot":{...} object with richer fields: sDesc/sCatalogDesc (HTML description), dHammerPrice, dHammerPremium, dEstimateLow/dEstimateHigh, images[]. The slug in the URL is not validated — any string works as long as auctionId/lotNo match, so a caller doesn't need Bonhams' exact descriptive slug.
  • robots.txt (checked 2026-09-05) disallows only /ldc/, /vms-assets/, *aggregate$, *head_image* — none of which this pack touches. No crawl-delay declared for User-agent: *.

Survey notes for the other two houses (Sotheby's, Phillips)

Bonhams was shipped first because its realized prices are plain SSR HTML with no gate — the easiest shape found. Quick probe notes on the other two, recorded for whoever builds their packs next:

  • Sotheby's — robots.txt sets Crawl-delay: 15 for User-agent: * and disallows /bsp-api/* (their actual API path) plus all PDF exports. Results pages are likely a genuine client-side API call behind /bsp-api/, which robots explicitly blocks — worth checking whether a different, non-disallowed endpoint serves the same data before assuming it's blocked outright.
  • Phillips — robots.txt disallows /Search, /search, /SEARCH, /bin/, /phillips/otis, /*/filter/. The obvious keyword-search path is blocked case-insensitively; a lot-detail page reached some other way (e.g. from a sitemap or a department listing) may still be fair game — not probed in this session.
  • Christie's — out of scope per the build plan (closed, internal API).

Bonhams proved the per-house-pack shape works, so follow-on fleet tasks were filed for the other two: #1280 (Sotheby's) and #1281 (Phillips), each carrying the probe notes above.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "bonhams": {
      "url": "https://gateway.pipeworx.io/bonhams/mcp"
    }
  }
}
What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/bonhams/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:

{
  "mcpServers": {
    "pipeworx": {
      "url": "https://gateway.pipeworx.io/mcp"
    }
  }
}

Both URLs reach the same gateway and the same 1875+ data sources. The only difference is which pack's tools are listed directly; ask_pipeworx reaches all of them from either one.

No MCP client? Call it over HTTP

curl -X POST https://gateway.pipeworx.io/v1/tools/bonhams_results_search \
  -H 'Content-Type: application/json' \
  -d '{"query":"Picasso"}'

No account needed for the first calls. Inspect any tool: GET https://gateway.pipeworx.io/v1/tools/bonhams_results_search. Find one: POST https://gateway.pipeworx.io/v1/tools/search_packs with {"query":"..."}.

Standalone (no gateway account)

This package also runs as a local stdio MCP server — no Pipeworx account, no gateway round-trip:

{
  "mcpServers": {
    "bonhams": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-bonhams"]
    }
  }
}

Or run it directly to confirm it starts:

npx -y @pipeworx/mcp-bonhams

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.

Using with ask_pipeworx

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:

ask_pipeworx({ question: "your question about Bonhams data" })

The gateway picks the right tool and fills the arguments automatically.

More

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.