Manufact's Launch HN Post Just Published the MCP Service Business Nobody's Selling Yet: Most Companies' MCP Servers Are Already Abandoned — Someone Has to Fix Them Before the App Stores Reject Them.

Manufact's Launch HN post is not really pitching a cloud platform.

It is publishing a diagnosis of every company's MCP server, in public, with a founder's own words.

Read past the pitch and you get admissions like these, straight from the post: "hardly anybody knows how to design a good MCP: most of them are 1:1 proxies of the API and are abandoned, since being one shotted a few months ago." And: "Auth is still a mystery for most teams (API key in the URL ???)." And: "Submission process on the stores is still quite tricky, manual and takes up valuable time."

That is not marketing copy. That is a list of paid engagements.

The business idea

Companies now feel pressure to have an MCP server — Manufact's post states it plainly: "all major tech companies have an MCP server." But having one and having a good one are different problems, and most companies only know how to build the former.

The service is simple: audit, redesign, and submission-readiness for company MCP servers, sold as a fixed-scope engagement to companies that shipped a thin wrapper months ago and haven't touched it since.

What the launch post admits is broken

Manufact's own founders wrote out the failure list for you:

• "Hardly anybody knows how to design a good MCP: most of them are 1:1 proxies of the API and are abandoned, since being one shotted a few months ago."
• "The MCP Spec advances quickly and it is not easy to keep track of the changes, and what they mean for your server."
• "Auth is still a mystery for most teams (API key in the URL ???)."
• "Most companies are not even aware that MCPs can return interactive UIs."
• "Clients still have to consolidate behavior, some do dynamic tool discovery, some don't, some persist authentication properly some don't."

Each bullet is a scoped deliverable. You don't need to guess what to sell — the launch post already wrote your service menu.

The number that sells the upsell

Manufact's post cites a concrete before/after: "an engineer at Amplitude reported that their MCP saw a 2x increase in retention after adding UI to their MCP."

That single sentence is the pitch for your highest-margin add-on. Most companies' MCPs are text-only tool calls. Turning one into an interactive UI — the kind MCP Apps (SEP-1865, official as of Jan 2026) now supports natively — is a scoped build with a real, citable retention number behind it.

Best customer profile

This works best for:

• Companies that already shipped an MCP server 6-12 months ago and haven't revisited it since — the ones most likely to be running an "abandoned," "one shotted" proxy
• SaaS products with an existing public API and real usage, who see MCP as the next required integration but have no one in-house who has actually designed a good tool interface
• Teams that got rejected (or are afraid of getting rejected) from the ChatGPT or Claude app directories and don't know why
• Companies with dashboards or visual data (analytics, ecommerce, reporting) who haven't realized their MCP could carry that UI directly into ChatGPT or Claude

How to package the offer

1. MCP audit

Score the client's existing server (or bare API, if they have no MCP yet) against Manufact's own failure list: is it a 1:1 proxy, does auth work cleanly, is there a UI opportunity, would it pass a store submission today.

2. Redesign and rebuild

Turn wrapper-style tool calls into a properly scoped tool set built around what users actually want to do — not what endpoints happen to exist.

3. Submission-readiness pass

Run the client's server through a pre-submission checklist before it ever touches ChatGPT's or Claude's actual review queue, so the first submission is the last one.

4. UI upgrade (upsell)

For clients with visual products, build the interactive UI layer and point to the Amplitude retention number as the business case.

5. Monthly spec-tracking retainer

MCP Apps just became an official extension in January 2026. The spec is still moving. Recurring revenue: keep the client's server compliant as the standard changes underneath them.

Why this beats generic "AI consulting"

Because the failure modes are not hypothetical — they are published, in the founders' own words, on the front page of Hacker News, describing the exact state of most companies' MCP servers today. You are not selling a vision of where AI is going. You are selling the fix for a problem a YC-backed infrastructure company just spent a launch post publicly diagnosing.

Bottom line

Manufact built the cloud. It did not build the redesign service for the thousands of companies whose MCP servers are already, in the founders' own words, abandoned proxies with mystery auth. That gap is a real, scoped, sellable engagement — and the launch post just did your discovery call for you.

Sources:

https://news.ycombinator.com/item?id=48762862

https://manufact.com