Hammerwerks ships with a built-in MCP server. Connect Claude Desktop, Cursor, ChatGPT (custom GPT), or any MCP-compatible client to your tenant. List auctions, approve lots, generate seller statements, query analytics — all from your assistant, with full audit-log visibility.
Full install guide →Read tools are safe; write tools are audit-logged on every call.
list_auctions — filter by statusget_auction — full detail + countslist_lots — browse per-auctionget_lot — photos, bids, AI confidencesearch_lots — full-text across tenantlist_bidders — KYC filterget_invoice — line items + paymentslist_invoices — per-auction or status filterget_analytics — sell-through, gross, top brandscreate_auction — new draft auctionapprove_lot — review → approvedupdate_lot — edit fields safelyrecord_payment — cash/check/wire logDeliberately NOT exposed: delete operations, refund-by-fiat, cancellations, KYC overrides. The MCP surface is for routine ops; destructive operations stay in the admin UI.
~/Library/Application Support/Claude/claude_desktop_config.json (or the Windows equivalent):{
"mcpServers": {
"hammerwerks": {
"transport": "http",
"url": "https://api.hammerwerks.com/mcp",
"headers": {
"Authorization": "Bearer hw_yourtenant_xxxxxx"
}
}
}
}list_auctions, returns the result.As of 2026, no other auction SaaS — Gavelry, HiBid, BidWrangler, NextLot, AuctionFlex — ships an MCP server. We do.
This matters most for the operators who run their shop on a MacBook and use Claude every day. Ask the assistant for analytics, have it batch-approve lots above a confidence threshold, generate a draft seller statement before you walk in to a consigner meeting. Every write is audit-logged in your tenant graph; you can see exactly what the assistant did, when, under which API key.