Hey IH community!
I've been building in the AI infrastructure space for the past year, and today I want to share what I've been working on.
In China, developers who want to use AI APIs face a frustrating reality:
A pay-per-call AI API marketplace with 67 endpoints covering OCR, speech recognition, text-to-speech, image generation, object detection, and more.
1. x402 Micropayments (USDC)
Instead of monthly plans, developers pay exactly for what they use — down to fractions of a cent. The x402 protocol handles payment negotiation automatically in the HTTP layer. No wallet top-ups, no minimum deposits.
2. Multi-payment support
Besides crypto, I integrated PayPal for international users and Alipay for Chinese developers. The goal: let developers pay however they want.
3. MCP Server
Published as goldbean-mcp on npm — developers can plug GoldBean's APIs directly into Claude, Cursor, or any MCP-compatible client. This was a game-changer for adoption.
Started as a side project to solve my own problem — I needed OCR and speech APIs for client projects but didn't want monthly commitments.
The hardest part wasn't technical. It was trust. Chinese developers are skeptical of new platforms. I focused on three things:
Developers hate pricing opacity. Every conversation with potential users comes down to the same question: "How much does this actually cost per call?" Monthly plans create anxiety. Pay-per-call creates trust.
Crypto payments are still a barrier. x402 is elegant, but many developers don't have USDC. That's why I added PayPal and Alipay — meet users where they are.
Distribution is harder than building. The product works, but getting developers to discover it is the real challenge. I'm currently exploring GitHub, npm, and developer communities.
I'd love feedback from this community on:
API: https://goldbean-api.xyz
GitHub: https://github.com/wuzenghai616-lang/goldbean
Thanks for reading!
i would prioritize time to first successful call over more endpoints. with 67 endpoints, a new developer may still ask which one to start with, what it costs, and whether the first response is reliable. pick three jobs such as OCR, speech, and image generation, give each a copy-paste example with the full price before execution, and show the same path through REST, SDK, and MCP. if a developer reaches a useful result in five minutes, that builds more trust than adding 33 more endpoints.
67 endpoints + micropayments — that's the infrastructure layer AI tools need. Most builders (me included) focus on the UI and forget the economics.
x402 is smart. Per-call billing beats subscription fatigue for dev tools.
The strongest part is the combination of pay-per-call pricing and payment flexibility. The 67 endpoints matter less than whether developers trust the marketplace enough to make it their default API layer.