Same request. Two completely different pipelines.

This walks through one sourcing query two ways: the way it's handled today, against five suppliers with five different systems — and the way it works once each supplier exposes an MCP server instead.

Get me the best price on 200 1/2" copper couplings, in stock, shipping this week

Illustrative simulation of a real sourcing workflow — timings are representative of the kind of gap teams report, not a benchmark of specific vendors.

Today — five bespoke integrations

0.0hrs
REST · Supplier A

Custom auth, custom schema. Someone already wrote this integration.

GET /api/v3/pricing?sku=CU-COUP-050 (proprietary auth header)
Scraping · Supplier B

No API. Price page layout changed last week — parser needs a fix before this works.

scrape_price_table(url) → selector mismatch
EDI · Supplier C

EDI 840/850 exchange. Batch process, response lands next business day.

EDI 840 sent → 850 pending
CSV · Supplier D

Download this week's price book, parse it, match SKUs by hand.

price_book_2026_09.csv (14,220 rows)
PDF · Supplier E
OCR the catalog PDF, hope the table extraction holds.
Manual

Normalize five different schemas into one comparable table.

Result after ~5 hours (one supplier still pending EDI response): 4 of 5 quotes in hand.

With MCP — one protocol, five servers

0.0sec
Connect

Agent connects to all five suppliers' MCP servers — same protocol, no custom code per vendor.

search_products()
search_products(sku: "1/2in copper coupling")
get_price()
get_price(sku, qty: 200) · get_bulk_discount(sku, qty: 200)
get_inventory()
get_inventory(sku) → in_stock, ships_by
Structured

All five return the same JSON shape. No parsing, no schema-matching.

Compare

Agent ranks all five on price, stock, and ship date automatically.

request_quote()
request_quote(supplier: "best_match", qty: 200)
Result: 5 of 5 quotes compared, cheapest in-stock option selected and quoted.

The pattern that actually matters: private B2B pricing

Most companies won't publish pricing publicly. The more durable pattern is an MCP server that sits in front of the ERP and returns the correct customer-specific price — without exposing the full price book to anyone.

Requester
AI agent
"Price this BOM for account #4471"
→
🔒
Auth-aware layer
Supplier's MCP server
get_customer_price(customer_id, sku)
get_contract_price(account)
create_quote(project)
estimate_shipping()
→
System of record
SAP / Oracle / NetSuite
(unchanged)

The MCP server doesn't replace the ERP or make pricing public — it's a consistent, permissioned interface the agent can query the same way it would query any other supplier, while the ERP still enforces who sees what.

Where adoption actually is today

Concentrated in software and data — not yet in physical goods or distribution, which is exactly where the opportunity sits.

TypeMaturityWhat's typical
AI model pricingMatureLive pricing exposed directly via MCP
SaaS pricingEmergingAgent/service marketplaces with MCP endpoints
Enterprise data productsEmergingData products and price tiers exposed via MCP
Physical product catalogsRareMostly pilots or custom builds
Manufacturer / distributor price booksVery rareStill REST, EDI, or private integrations — this is the gap

MCP doesn't replace the ERP. It becomes the AI-facing interface to it — the same role a website played for browsers, now for agents. Companies that wrap their existing price books this way get found, compared, and quoted by AI automatically; the ones that don't stay invisible to it.