altfinance-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@altfinance-mcp find the top 10 most expensive Rolex watches sold after 2020"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ALT/FNDATA MCP Server
Connect your AI assistant to luxury and collectibles auction data. Query watches, jewelry, loose gemstones, handbags, cars, motorcycles, aircraft, wine and whisky, fine art, works of art, design, fashion, shoes, automobilia, books, coins, perfume, accessories, musical instruments and other vehicles from 100+ global auction houses, in plain language.
It uses the open Model Context Protocol (MCP), so it works with Claude, ChatGPT, Gemini, Perplexity and other assistants and developer tools that support MCP servers.
Quick Start
1. Get an API Key
Get your key from API Key & MCP in the ALT/FNDATA app.
2a. Connect the hosted server (no install)
In any assistant that supports remote MCP servers (often called custom connectors), add a connector with this URL:
https://mcp.altfndata.com/mcpChoose OAuth if you are asked for an authentication method. A sign-in page then asks for your ALT/FNDATA API key.
Claude: Settings → Connectors → add a custom connector.
ChatGPT: add it as a custom connector (developer mode). ChatGPT only connects to hosted servers, so use this option.
Perplexity: Settings → Connectors → add a custom connector with the URL above.
Gemini app: Settings → Connected Apps → add a custom app with the URL above (personal Google accounts).
Gemini Enterprise: an admin adds ALT/FNDATA once for the organisation. See Gemini Enterprise set-up below.
Menu names and plan availability are set by each assistant and change over time; check its help pages if the steps differ.
Gemini Enterprise (admin set-up)
Gemini Enterprise connects through a Custom MCP Server data store, which an admin sets up once for the organisation.
Register a client. The response contains a
client_idand aclient_secret. The secret is shown only once, so store it safely.curl -X POST https://mcp.altfndata.com/register \ -H "Content-Type: application/json" \ -d '{"client_name": "Gemini Enterprise (your organisation)", "redirect_uris": ["https://vertexaisearch.cloud.google.com/oauth-redirect"], "token_endpoint_auth_method": "client_secret_post", "grant_types": ["authorization_code", "refresh_token"], "response_types": ["code"], "scope": "read"}'In the Gemini Enterprise console, add a Custom MCP Server data store with these values:
Field
Value
MCP Server URL
https://mcp.altfndata.com/mcpAuthorization URL
https://mcp.altfndata.com/authorizeToken URL
https://mcp.altfndata.com/tokenClient ID
client_idfrom step 1Client Secret
client_secretfrom step 1Scopes
readEnable PKCE
ticked
Each user then connects from Gemini and signs in with their own ALT/FNDATA API key. The client ID and secret only identify your organisation's Gemini Enterprise; they give no access to data on their own.
Google's guide to setting up a custom MCP server describes the console screens.
2b. Or run it locally
For desktop apps and developer tools that launch MCP servers on your machine, either install it with pip:
pip install altfinance-mcpor run it with no install using uv — point the client's command at uvx with altfinance-mcp as the argument (see the uvx config example below):
uvx altfinance-mcpMost tools take a configuration in this shape; add it to the tool's MCP settings and restart it.
Example, Claude Desktop: the config file is %APPDATA%\Claude\claude_desktop_config.json on Windows and ~/Library/Application Support/Claude/claude_desktop_config.json on macOS.
{
"mcpServers": {
"altfinance": {
"command": "altfinance-mcp",
"env": {
"ALTFINANCE_API_KEY": "your-api-key-here"
}
}
}
}With uvx (no pip install needed), use "command": "uvx" and pass the package as an argument:
{
"mcpServers": {
"altfinance": {
"command": "uvx",
"args": ["altfinance-mcp"],
"env": {
"ALTFINANCE_API_KEY": "your-api-key-here"
}
}
}
}Example, Claude Code:
claude mcp add altfinance -- altfinance-mcpSet the env var before launching:
export ALTFINANCE_API_KEY="your-api-key-here"Related MCP server: Catawiki MCP Server
What You Can Ask
Once connected, ask your assistant questions like:
"What tables are available in the ALT/FNDATA database?"
"Show me the schema for the watches table"
"Find the top 10 most expensive Rolex watches sold after 2020"
"How many Hermes handbags were sold by Christie's?"
"Give me market stats for Patek Philippe watches sold at Phillips"
"How have Hermes Birkin prices moved over the last five years?"
"Price-check a Rolex Daytona 116500 white dial"
"Search the whole market for a Burma ruby ring over $50k"
"What loose sapphires have sold at auction?" (uses
gems, notjewelry)
Available Tools
All 9 tools are read-only and annotated with readOnlyHint: true. No data is modified by any tool.
Tool | Description |
| Lists the 20 available tables with live descriptions and column counts |
| Shows column names and types for one table |
| Queries a table with filters, sorting, field selection, and pagination |
| Realized-price summary for a table or filtered segment: count of sales, median, 25th/75th percentiles, average, min/max, sale-date span, and top vendors by volume |
| How realized prices moved over time, bucketed by year, quarter, or month (count, median, average per period) |
| Fair-value range for a single item from comparable sales, with source links. Live comp categories: handbags, watches, cars |
| Whole-market keyword search across every category at once, merged by realized price. With no category it fans out across all 20 tables |
| Precomputed model-level price index (low / 25th / median / 75th / high, sample size, trend) for a brand or model. Covered categories: handbags, watches, cars |
| Read-only view of price alerts, or guidance on enabling them. See "Price alerts are read-only" below |
Every tool carries readOnlyHint and openWorldHint.
Price alerts are read-only
Price alerts belong to a signed-in ALT/FNDATA account (Cognito login), not to an API key. The MCP authenticates with an API key only, so it cannot create or enable alerts, and the alerts feed itself is not reachable for an API-key caller. list_alerts therefore either reports fired alerts (when a session makes the feed reachable) or returns guidance on setting alerts up in the app at app.altfndata.com. Alert allowance by plan: free 0, registered 1, collector 10, pro 100, included with Membership.
Feature and plan suggestions
Some tool responses may append a single short "💡" line suggesting a complementary ALT/FNDATA product or a plan upgrade. This happens at most once per response, only at a genuine tier boundary (a table your plan does not include, or an exhausted row quota) or as a relevant feature tip, and never for unlimited or comp accounts. It never changes the data returned.
Query Capabilities
Filter by any column using operators:
eq,neq,gt,gte,lt,lte,like,in,not_in,is_null,is_not_nullSort by any column, ascending or descending
Select specific fields to return
Paginate with limit/offset (max 500 rows per request)
Data Coverage
The database spans 20 tables. Call list_tables for the live set with current column counts.
Table | Description |
| Luxury watch auction results |
| Luxury handbag auction results |
| Finished jewelry, mounted pieces (rings, necklaces, bracelets, earrings) set with gemstones |
| Loose, unmounted gemstones (single stones and parcels). Use this for gemstone valuations, not |
| Collector car auction results |
| Collector motorcycle auction results |
| Aircraft auction results |
| Wine and whisky auction results |
| Works of art with stone/jewelry relevance (jade, jadeite, diamond-set, etc.) |
| Fine art auction results across all movements and periods |
| Design and decorative arts: furniture, lighting, glass, ceramics, studio/industrial design |
| Fashion and couture: clothing and accessories (not handbags or shoes, which have their own tables) |
| Footwear: designer shoes and collectible sneakers |
| Automotive memorabilia, parts, signage, models and related collectibles (not the cars themselves) |
| Books, manuscripts, maps and printed material |
| Numismatic coin auction results |
| Other vehicles: boats, tractors, trucks, buses and military vehicles (not cars or motorcycles) |
| Perfume and collectible fragrance bottles and flacons |
| Fashion accessories: belts, scarves, cufflinks, brooches and sunglasses, including gem-set pieces |
| Musical instruments: guitars, violin family, pianos and keyboards, woodwind, brass, percussion and amplifiers |
gems vs jewelry: gems holds LOOSE, unmounted stones (the right table for a gemstone valuation); jewelry holds FINISHED, mounted pieces set with stones. If one returns nothing, try the other.
Data sources include Christie's, Sotheby's, Phillips, Bonhams, Artcurial, Dorotheum, Heritage Auctions, and 100+ other auction houses worldwide.
Hosted Server
The hosted server at https://mcp.altfndata.com/mcp (Quick Start 2a) is this package running in HTTP mode. Any MCP client that supports remote servers with OAuth can connect to it; the sign-in page asks for your ALT/FNDATA API key, so no local installation is needed.
Self-Hosting the Remote Server
To run the MCP server in remote (HTTP) mode:
export ALTFINANCE_TRANSPORT=streamable-http
export ALTFINANCE_ISSUER_URL=https://mcp.altfndata.com
export ALTFINANCE_RESOURCE_URL=https://mcp.altfndata.com
python -m altfinance_mcpOr with Docker:
docker build -t altfinance-mcp .
docker run -p 8000:8000 \
-e ALTFINANCE_TRANSPORT=streamable-http \
-e ALTFINANCE_ISSUER_URL=https://mcp.altfndata.com \
-e ALTFINANCE_RESOURCE_URL=https://mcp.altfndata.com \
altfinance-mcpThe remote server exposes:
POST/GET/DELETE /mcp— MCP Streamable HTTP endpoint (authenticated)GET /.well-known/oauth-authorization-server— OAuth metadataPOST /register— Dynamic Client RegistrationGET /authorize— OAuth authorizationPOST /token— Token exchangeGET/POST /login— API key entry form
Environment Variables
Local (stdio) mode
Variable | Required | Default | Description |
| Yes | — | Your ALT/FNDATA API key |
| No |
| API base URL |
Remote (HTTP) mode
Variable | Required | Default | Description |
| Yes |
| Set to |
| No |
| API base URL |
| No |
| HTTP listen address |
| No |
| HTTP listen port |
| No |
| OAuth issuer URL (your public URL) |
| No | Same as issuer | OAuth resource server URL |
Privacy Policy
This MCP server connects to the ALT/FNDATA API (api.altfndata.com) to retrieve auction data. The following outlines our data practices:
Data collection: The server sends your API key and query parameters (table names, filters, sort options) to the ALT/FNDATA API. No personal user data is collected beyond what is required for API authentication and query execution.
Usage and storage: Query data is processed in real-time and returned to the MCP client. The server does not persist, cache, or store any query results or user data locally.
Third-party sharing: Data is exchanged only between the MCP client and the ALT/FNDATA API (
api.altfndata.com). No data is shared with any other third parties.Data retention: The server is stateless. No user data or query history is retained after a session ends.
Contact: For privacy inquiries, contact support@altfndata.com.
Full privacy policy: https://altfndata.com/privacy-policy/
Support
Documentation: https://docs.altfndata.com
Email: support@altfndata.com
Requirements
Python 3.10+
An ALT/FNDATA API key
License
MIT
Available Tools
9 toolsget_table_schemaARead-onlyInspect
List a table's column names and types so you can build a precise query.
Call this before query_table (or before adding filters to market_stats /
price_trend) to learn the exact field names a table exposes — e.g. whether
prices are in usd_price_decimal, how the brand/designer field is named, or
whether a carats column exists. Avoids guessing column names.
Example: before querying Rolex sales, call get_table_schema("watches") to
confirm the brand field is designer and the price field is
usd_price_decimal.
Args: table: Table alias (call list_tables for the full set). Note: "gems" = loose/unmounted stones; "jewelry" = finished/mounted pieces.
Inputs: table (required). Returns: result
list_alertsARead-onlyInspect
Show ALT/FNDATA price alerts (read-only) or explain how to set them up.
Price alerts watch a saved item and fire when a matching lot sells at or below your target. They belong to a signed-in ALT/FNDATA account, not to an API key — so alerts are created and managed in the app (app.altfndata.com), and this MCP can only read/report them. If your key isn't tied to a signed-in session, this returns guidance on enabling alerts (allowance: free 0 · registered 1 · collector 10 · pro 100 · included with Membership) rather than a data feed.
Args: limit: Max fired alerts to show if the alerts feed is reachable (default 25).
Inputs: limit. Returns: result
list_tablesARead-onlyInspect
List every auction-data table ALT/FNDATA serves, with live descriptions and column counts.
Start here whenever a question names a category and you need the right table
alias to pass to get_table_schema, query_table, market_stats or price_trend.
The set spans watches, jewelry (finished/mounted pieces), gems (loose
stones), handbags, cars, motorcycles, wine & whisky, works of art, fine art,
coins, and more. Always call this for the current set rather than assuming
it, and note that loose gemstones live in gems, not jewelry.
Example: for "what categories of data do you have?" or before "show me the
most expensive coins sold at auction" (to confirm the coins alias exists).
Inputs: no named parameters. Returns: result
market_indexARead-onlyInspect
Precomputed model-level price index (headline figures) for a model or brand.
Reads the ALT/FNDATA Index's precomputed hot-stats — instant, no scan. Give a
typed model query (e.g. "Hermès Birkin 25", "Rolex Daytona 116500") for a
single model's realized-price band (low / 25th / median / 75th / high), sample
size, and recent trend; or give a brand alone for that brand's whole set of
model indices. Covered categories are handbags, watches, and cars. For other
categories, use market_stats (whole-market read) or price_trend instead — this
tool never invents a figure that isn't precomputed.
Args: query: A specific model to look up (brand + model + optional size/material). brand: A brand to list all model indices for (alternative to query). category: One of "handbags", "watches", "cars". Inferred from the text when omitted.
Inputs: brand, query, category. Returns: result
market_statsARead-onlyInspect
Summarize the realized-price market for a table or a filtered segment.
Returns count of sales, median, 25th/75th percentiles, average, min/max of
usd_price_decimal, the sale-date span, and the top vendors by volume — all
computed server-side over the full matching set (not a sample).
Args: table: Table alias (call list_tables first). "gems" = LOOSE, unmounted stones (use for a gemstone-market read); "jewelry" = FINISHED, mounted pieces. A loose-stone question can span both. filters: Optional list of filter dicts, same shape as query_table (field/op/value). Narrow to a segment, e.g. [{"field": "designer", "op": "eq", "value": "Patek Philippe"}]. sold_only: When true (default), restricts to sold lots with a positive USD price so the stats reflect realized transactions.
Inputs: table (required), filters, sold_only. Returns: result
price_checkARead-onlyInspect
Estimate a single item's fair-value range from comparable auction sales.
Give either a free-text description (e.g. "Hermès Birkin 25 Togo, gold hardware") OR a lot URL from an auction house. Returns a realized-price range (low / median / high), a confidence read, and comparable sales with source links. Best for valuing ONE specific piece; for a whole-market read use market_stats.
Live comp categories are handbags, watches, and cars. Loose gemstones and
jewelry are intentionally held from comp-matching (under-specified colored
stones give unreliable medians) — for those, use market_stats on gems
(loose stones) or jewelry (mounted pieces) instead.
Args: description: Free-text item description. lot_url: A lot URL from an auction house (alternative to description). category: One of "handbags", "watches", "cars". Inferred from the text if omitted.
Inputs: lot_url, category, description. Returns: result
price_trendARead-onlyInspect
Show how realized prices moved over time for a table or filtered segment.
Buckets sold lots by sale period and reports, per period, the number of
sales, the median, and the average usd_price_decimal.
Args: table: Table alias (call list_tables first). Gems = loose stones; jewelry = mounted pieces. filters: Optional filter dicts (field/op/value), same shape as query_table. by: Time bucket — "year" (default), "quarter", or "month". sold_only: Restrict to sold lots with a positive USD price (default true).
Inputs: table (required), by, filters, sold_only. Returns: result
query_tableARead-onlyInspect
Query one known table with precise filtering, sorting, and field selection.
Use this when you already know the target table (from list_tables) and want
exact control — e.g. "the 10 most expensive Rolex watches sold since 2020"
becomes table="watches", filters on designer/sale_date, sort by
usd_price_decimal desc, limit 10. If you don't know which table holds the
item, use search (whole-market) instead; for aggregate figures (median,
percentiles) use market_stats.
Args: table: Table alias (call list_tables for the full set). "gems" = loose stones (use for gemstone valuations); "jewelry" = finished pieces. fields: List of column names to return (e.g., ["item_title", "vendor", "usd_price_decimal"]). If not specified, returns all columns. filters: List of filter conditions. Each filter is a dict with: - "field": column name - "op": operator — one of "eq", "neq", "gt", "gte", "lt", "lte", "like", "in", "not_in", "is_null", "is_not_null" - "value": the value to compare against Example: [{"field": "vendor", "op": "eq", "value": "Phillips"}, {"field": "usd_price_decimal", "op": "gte", "value": 50000}] sort_field: Column name to sort by (e.g., "usd_price_decimal" or "sale_date") sort_direction: "asc" or "desc" (default: "desc") limit: Maximum rows to return (default: 20, max: 500) offset: Number of rows to skip for pagination (default: 0)
Inputs: table (required), limit, fields, offset, +3 more. Returns: result
searchARead-onlyInspect
Whole-market keyword search across every ALT/FNDATA category at once.
Runs a server-side free-text search (the same engine behind the ALT/FNDATA
Search app): your query is split into tokens, each matched across
item_title, designer, vendor, gem_type, primary_gem, origin, color, model
and sale_title with synonym expansion (e.g. ruby↔rubis, jade↔jadeite). With
no category it fans out across the whole market and merges the top realized
sales by USD price — so you don't have to guess which table holds an item.
Args: query: Free-text keywords, e.g. "ruby ring", "Daytona 116500", "Birkin 25". category: Optional hint to narrow the search — a table alias (e.g. "watches") or a plain word ("cars", "gemstone"). A gemstone/stone hint searches BOTH gems (loose) and jewelry (mounted); omit it to search everything. filters: Optional extra filter dicts (same field/op/value shape as query_table), AND-ed with the keyword match — e.g. [{"field": "usd_price_decimal", "op": "gte", "value": 50000}]. limit: Max merged results to return (default 25, max 100).
Inputs: query (required), limit, filters, category. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 9 tools
Most tools occupy distinct slots: list_tables (discovery), get_table_schema (introspection), query_table (precise rows), market_stats (aggregates), price_trend (time series), price_check (single-item comp), search (cross-market keywords), market_index (precomputed model index), list_alerts. The only soft boundaries are market_stats vs market_index (segment stats vs precomputed model band) and query_table vs search, but descriptions explicitly tell the agent which to pick and when to fall back.
All names are snake_case, but conventions are mixed: verb_noun (list_tables, get_table_schema, query_table, list_alerts), noun_phrase (market_stats, price_trend, price_check, market_index), and a bare verb (search). The market_* family is internally coherent, yet the surface as a whole does not follow one predictable pattern, so an agent can't infer a namespace purely from a name.
Nine tools for an auction-market data server is well scoped: discovery, schema, row query, aggregate, trend, valuation, search, index, and alerts each cover a genuinely different need. Nothing looks redundant or padded, and there is no obvious missing layer that would require an additional tool.
The read-only lifecycle is well covered end to end: find tables, inspect schema, query rows, aggregate, trend, value a piece, and search across categories. Gaps are minor but real — no single-lot lookup by ID/URL independent of price_check, no cross-table aggregation across categories (e.g. gems+jewelry), and alerts are read-only by design, though each gap is documented with a workaround.
Maintenance
Related MCP Connectors
Collector-car auction history, live listings, bid trails, and market statistics for AI agents.
CompSniper gives AI assistants real eBay sold prices: up to 240 sold listings per call, 3 years (1,095 days) of eBay sold history on every plan including free, and the real accepted Best Offer price (what the buyer actually paid). 18 marketplaces on one key: 8 eBay sites (US, UK, Germany, France, Italy, Spain, Canada, Australia) plus TCGplayer, PSA, Grailed, Poshmark, Mercari, Depop and the Goldin, Fanatics Collect, Pristine Auction and MySlabs auction houses. Results come raw or AI-cleaned (accessories, lots and wrong models removed), with median, range and shipping. Free plan: 100 requests a month, no card. In the Claude Connectors Directory and ChatGPT Plugins.
AI-native art catalogue. Catalogue works, parse provenance, and generate signed RAIs.
The agent-native marketplace for AI data: search listings and read their price and rights.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceProvides access to government auctions data for physical-asset auctions including surplus, seized, and forfeited items. Designed for AI agents to query and retrieve auction information via natural language.132 npm2MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to read-only access Catawiki auction marketplace data, including searching lots, inspecting items, browsing categories, and exploring auctions.MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI agents to query CS2 item market analytics, including historical K-lines, live broad index trends, price snapshots, float/wear parsing, and cloud-rendered 3D inspection images.MIT
- AlicenseNot gradedqualityBmaintenanceEnables read-only querying of luxury auction results across twenty categories, including filtered table searches, market statistics, price trends, fair-value checks, whole-market search, brand/model price indexes, and price alerts.335 npmMIT