I’ve watched buyers send the same request for quotation to twelve factories and hear back from three, two weeks later, after the deal was already done. The problem isn’t the factories. It’s that the search for suppliers is moving faster than they can respond. By the end of 2026 the buyer at the other end of a quote request will increasingly not be a person but an AI assistant working for one. ChatGPT, Claude and specialised services like GETMILK can already ask suppliers questions, compare quotes and assemble production plans on a buyer’s behalf. They do it through one shared standard called MCP (Model Context Protocol). A factory that cannot be asked through it today is where a hotel without a website was in 2002: present in the world, invisible to the channel that’s taking over how customers find you.
This post explains, for owners and managing directors, why buyers’ AI assistants will look for suppliers this way, what an MCP server does for a factory in terms of sales, cost, risk and control, and what to do next. Examples come from the 2,000+ manufacturers in the GETMILK index (Portugal mapped in depth, the rest of Europe following).
1 · What MCP is, in one paragraph
MCP is an open standard that Anthropic released in November 2024 and handed to the Linux Foundation’s Agentic AI Foundation in December 2025. It is now supported by the major AI providers, Anthropic, OpenAI, Google and Microsoft among them. In plain terms, an MCP server is a small, secure front desk for your company that AI assistants can talk to. It lets them do three kinds of things, and only the ones you allow:
- Read what you publish: your profile, processes, materials, machines, certifications and current lead times. Think of it as your catalogue, written so software can understand it.
- Ask you to act: check whether a part fits your capabilities, give a price band, or register a quote request with your sales team.
- Follow your house rules: how you like quote requests formatted, your design-for-manufacturing checklist, which material you suggest when the one asked for is slow.
The effort is similar to building a small website feature. The difference is the audience: a website is read by people, an MCP server is read by the assistants those people now delegate to. We explain how it connects to your ERP on the MCP and ERP integration page.
2 · Why a manufacturer should have one
Buyers are handing supplier search to AI assistants
When a hardware founder needed a CNC shop in 2020, they Googled and emailed. In 2026 they ask their assistant. It lists options, asks each one, and returns a ranked shortlist. If you’re an Iberian foundry with a 1998 website and nothing an assistant can read or ask, it skips you, even if you’re objectively the best fit. Being reachable is your ticket onto that shortlist.
Answer in seconds, not days
Most factories lose deals on response time. A buyer sends a drawing at 14:00 on Tuesday; if you reply on Friday with “we can do it for €X with a 6-week lead time”, the buyer has already signed with a supplier whose system answered in 30 seconds. A quoting function linked to your ERP, or even just to a well-calibrated cost calculator, closes that gap. You stay in the running.
You decide what the world sees
When third parties copy information from your brochures and PDFs, they choose what to show. An MCP server is your own statement. Public capabilities stay public; price lists are only shown to buyers who have identified themselves; trade secrets never leave your systems. You set the level of detail.
A level field against larger players
A 25-person Portuguese precision-machining shop normally can’t outspend a German tier-1 on marketing. But to an AI assistant, both are just answers to the same questions. If you honestly answer “3 weeks” while the tier-1 answers “12 weeks”, the assistant can rank you higher on lead time. That’s an advantage you don’t get on a trade-show floor.
A possible new revenue line
MCP has no payment system built in, but every request to a remote server comes from an identified caller, so usage can be counted and charged. Once your server is the normal way buyers reach you, a pay-per-qualified-request model becomes natural (€2–€10 per qualified request is our working assumption, not a market figure). This is new revenue, not a charge on existing customers.
3 · What to offer, and where to start
Based on the 27 manufacturing processes and 1,000+ profiles GETMILK has indexed, a useful first version lets an assistant:
- Read your company profile and languages, processes with tolerances and part sizes, materials in stock, machine list, current lead times by process, certifications (ISO 9001, AS9100, IATF 16949 and others), public references, and commercial terms such as minimum order, NDA and payment terms.
- Ask whether a part fits (yes, partly or no, with a reason), get a price band with lead time and validity date, send a formal quote request to a named person with a response promise, hold a tentative capacity slot, request a sample, and get design-for-manufacturing feedback.
4 · Costs and risks, honestly
- Effort. Most of the work is tidying data you already have: a capability list in a Word document, a machine list in Excel, lead times in a sales rep’s head. Very little new data has to be created.
- Wrong answers. A price band that is out of date is worse than none. Every answer should carry its assumptions and a validity date, and a person on your side confirms before anything becomes a commitment.
- Exposure of sensitive data. Keep the connection read-only against a copy of the relevant ERP data, show prices only to identified buyers, and keep formulas and margins out entirely. See MCP and ERP integration for how we set this up.
- Doing nothing. The cost of staying invisible is the quotes you never hear about, and it grows as more buyers delegate the search.
5 · How GETMILK uses your server
GETMILK takes a buyer’s request through three stages: analyse (turn whatever the customer sent into a clear product brief), breakdown (match it to processes, materials and suppliers) and plan (a 90-day action plan with budget and contacts). Suppliers with an MCP server change the breakdown from “here are some EU manufacturers who’ll probably fit” to “Polopiqué has 800 kg of Nm 30/2 cotton in stock, can spin from week 18, €5.40/kg at this volume.” (illustrative figures, not a real quote).
- A customer drops a spec on getmilk.tech: a sketch, photo, link or a paragraph of notes.
- GETMILK turns it into a structured brief (processes, materials, tolerances, target volume).
- It finds candidates in its index of 2,000+ shops (Portugal-first) by process and country preference.
- Every candidate with an MCP server is asked at once whether the part fits; a ten-supplier check should take seconds and separates real fits from nominal ones.
- The top three fits per process are asked for a price band, lead time and validity date, stamped by the manufacturer’s own system.
- The plan says who to contact, what to send, when to expect the first article and what it costs.
- When the customer presses “Submit RFQ”, each chosen manufacturer receives a complete, pre-formatted request in their usual inbox or CRM: no copy-paste, no language barrier, no missing fields.
6 · Example: TEXFIO conductive yarn
Take a textile spec recently launched on the platform: TEXFIO, washable PEDOT:PSS conductive yarn for smart apparel. The customer wants 50 kg of conductive yarn coated for at least 50 wash cycles, base yarn Nm 30/2 cotton.
Without MCP, GETMILK returns a thoughtful but vague plan: “Riopele or Polopiqué likely fit; expect ~€5–7/kg; ~12 weeks lead time.” Useful, not actionable.
If each Vale do Ave mill ran a server, the answers would look like this (illustrative, the numbers are invented to show the shape):
- Polopiqué: partial fit, spinning yes, coating through partner CeNTI; €5.40/kg, 8 weeks, base yarn from stock, valid until 31 October 2026.
- Inovafil: €4.95/kg, 11 weeks.
- Riopele: no fit, weaving and finishing only, no spinning.
GETMILK then writes a sourcing plan that names the mills with their own numbers, not generic ranges, and an honest “no” from the one that does not spin. The customer picks the fastest option and presses “Submit RFQ”; Polopiqué’s sales lead has a complete request in their CRM 90 seconds after the customer first dropped a paragraph of notes on getmilk.tech.
7 · Where this lands first in Europe
- Vale do Ave (Portugal): textile spinning, knitting, dyeing, finishing. Dozens of mid-sized mills already on ERPs such as PHC Software or Primavera, which could add an MCP layer in a week.
- Lombardia and Veneto (Italy): precision CNC, sheet metal, leather goods. German tier-1s already buy here; MCP would let smaller buyers in.
- Baden-Württemberg (Germany): mostly automotive-grade CNC and electronics assembly. Slowest to adopt because of existing electronic ordering links, but the largest deals once they do.
- Stara Zagora and Plovdiv (Bulgaria), Łódź (Poland): emerging composites, drone parts and electronics assembly. Hungry for this because they don’t have sales teams to chase buyers across the EU.
What to do next
- Gather what you already have. Capability list, machine list, certifications, typical lead times. This is most of the work.
- Decide what is public and what is not.Capabilities public, price bands for identified buyers only, margins and formulas never.
- Connect it to your source of truth. If you have an ERP, even an older PHC or Primavera install, the server reads a safe copy of the relevant data. If you don’t, a shared spreadsheet is enough for a first version. Your IT partner or ours can do this; see MCP and ERP integration and AI agents.
- Make it findable. List it in the official MCP Registry, mention it on your website and send it to us; we add it to your GETMILK profile.
- Watch what buyers ask. Within a week you will see which questions matter and where to invest next.
The web reorganises roughly once a decade. Search reorganised it around pages, mobile around apps; AI assistants are reorganising it around questions a supplier can answer directly. For a manufacturer, that is the difference between being a name an assistant lists and a name it has never heard of. The work to close that gap is small; the cost of staying on the wrong side of it grows every week.