Skip to content

Versum MCP

Every local business already publishes its prices. The problem is that no agent can read them, so the answer to "what does that shop charge" is still a phone call at eleven in the morning.

Server type: Remote (Streamable HTTP)
Endpoint:    https://versum.suverse.io/mcp
Auth:        none for reads
Tools:       13 measured, all read-only
Coverage:    128 trades, 5,398 cards, 9,300 businesses
Writes:      none. No order is placed, nothing is reserved

What it does

Versum is a free index of real local businesses in the United States, built by reading each business's own public website rather than a directory feed. The point is not the listing. The point is the shelf: what each business sells, at what published price, with the link that goes back to the shop.

The thirteen tools split by the question being asked.

Question Tool
What trades exist, and how deep is each one list_niches
Surf the shelf list_cards, search_catalog
The user named a business find_business
One business in full get_card, get_business
One item, with a live price request get_offering, get_product
What would this cost delivered quote_order
Stock check_availability
Act place_order, book_slot, order_status

get_product and get_business are synonyms of get_offering and get_card, kept so an agent that already knows the other name does not have to relearn it.

The contract is written for an agent rather than for a human reader, which is the part worth copying. Ask search_catalog for pizza without a location and it refuses to guess: needs.missing: location, a stated reason that no distance was measured so nothing was narrowed, and an instruction to ask the user for a city or a 5-digit postcode before quoting any of it back. It also states what the answer is not: the order of the results says nothing about proximity.

What this pass measured

A keyless session against the live endpoint, not a documentation read.

initialize answered HTTP 200 with serverInfo versum-catalog 1.6.0 on protocol 2025-06-18, tools/list returned all 13 tools with schemas in 50,179 bytes, and no credential was used at any point.

list_niches returned the whole index in one object: 128 trades, 5,398 cards and 9,300 businesses, with per-trade coverage and the city zones behind it. The depth is uneven and the index says so. Restaurant carries 783 cards over 1,042 businesses, coverage partial, with the reason printed as Tampa, FL: 10 cell(s) still answer the cap. Florist is 183 cards over 316 businesses and complete. Pizza is 95 over 202 and complete. Each zone carries the date it was last passed, 2026-11-01 and 2026-11-02 in the samples read.

search_catalog for pizza in Tampa returned 12 matches in a 15 mile radius and the first three businesses with their items. The top card is a real shelf with real prices:

  • Green Bench Brewing Co., 0.3 miles away, (813) 867-8804, platforms skytab and squarespace, checked: 2026-11-01
  • FOUR CHEESE PIZZA at $12.95, price_type: fixed, availability: unknown
  • Pepperoni Pizza at $14.50, same terms

get_offering on the first item returned the price as a claim with its evidence attached rather than as a number: price_state: {state: confirmed, basis: published, source: semantic_html, read_at: 2026-11-01T01:28:53}, evidence: {url: http://thepintandbrew.com/menu, quote: "$ 12.95"}, currency: USD. The stock line is the honest half: availability_state: {state: unknown, basis: not published}, and the card says availability not published for any offering shown. Somebody can be told what a pizza costs and cannot be told whether it is in stock, and both facts arrive labelled.

get_card returned the full business: address, latitude and longitude, e-mail, website, platform squarespace, timezone America/New_York, hours for all seven days as 11:30 to 00:00 with a quote field written as a person would say it, and niche_decided_by: {source: catalog, reason: "catalog: 100% restaurant (dish 25, drink 8) of 33 classified, 46 on the shelf"}. The classification carries its arithmetic, so a trade assignment can be argued with instead of trusted.

find_business is the operational tool. Asked for Green Bench Brewing near 33602 with intent: order and needed_by: 2026-11-06T19:00:00-05:00, it read the timestamp as open_at on Fri 2026-11-06, matched at 0.9 on part, and returned a task card: the phone number, the hours for that day with state: confirmed and the reasoning open at 19:00 by the published hours (11:30 AM - 12:00 AM), and the five cheapest items with prices, Bottle Water at $3.00, Iced Tea at $3.95, Lemonade at $3.95 and two more. That is the whole answer an operator needs before sending someone: open then, this is what it costs, here is the number.

quote_order on the $12.95 pizza delivered to postcode 33602 came back placed: false, status: not_available, reason: no_public_cart, and stated its own limits: the shop's public cart priced the order in a session of its own which was emptied afterwards, only the delivery address was sent, no name, e-mail or phone, and nothing about it was kept. The quote path failing is worth knowing before anybody builds on it: this shop publishes a menu but not a cart.

Where it fits, and where it does not

Nothing else in this catalogue reads a local shop's own shelf. GBPly MCP manages your own Google Business Profile and keeps every write behind an approval that expires in fifteen minutes; Listings API MCP publishes a business out to 80 plus directories and maps; GoodLeads MCP turns new-business state filings into priced lead lists. All three act on business identity. None answers what a shop on the next block charges for a pizza. Undercart MCP is the nearest neighbour and covers online Shopify stores; this covers the shop with a Squarespace menu and a phone number.

The limit is stated by the server itself rather than by this guide. Coverage is partial and city-bounded: Tampa is deep, and a query in a city with no passed cells has nothing to return. There is no write path a buyer should rely on until a card shows a connected platform, and the one shop measured here has a menu but no public cart. The index is US only.

Integration notes for CorpusIQ

  • Add it as a read-only source beside a commerce or Shopify connector and the answer stops being a link: a price, a distance, a phone number and an honest unknown.
  • find_business with needed_by is the pattern worth reusing on our own side. A timestamp that resolves to open_at with the reasoning printed is a citation an operator can check, which is the difference between an answer and a guess.
  • No key, no account, nothing to configure. The call that costs nothing is also the one that says what it does not know.
  • Server: https://versum.suverse.io/mcp
  • Website: https://versum.suverse.io
  • Public design notes: https://github.com/sudzikcoin/versum-agent-assistant
  • Registry: io.github.sudzikcoin/versum-agent-assistant

Verification

  • Endpoint driven keyless on 2026-11-02: initialize 200, versum-catalog 1.6.0, 13 tools, 50,179 bytes of schema.
  • Six tools called live: list_niches (128 trades, 5,398 cards, 9,300 businesses), search_catalog (12 matches, 3 cards), find_business (task card with a confirmed open state), get_card, get_offering (price with evidence quote, stock unpublished and labelled), quote_order (placed: false, no_public_cart).
  • Two argument-shape corrections found by driving it: get_card and get_offering take id, not business_id or offering_id, while quote_order and place_order take offering_id and business_id. The schema is strict, additionalProperties: false, and a wrong field name returns JSON-RPC -32602.
  • Numbers in this guide are what the server returned on the day, not the counts in the submission issue. The issue states 7,415 cards over 128 trades; the live list_niches answered 5,398 cards over 9,300 businesses across the same 128 trades.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing