I have 61 MCP servers and Chrome extensions, all with Stripe payment links baked in.
No, I didn't write them all by hand. I built a pipeline where AI agents ship products end-to-end — from GitHub repo to npm to Smithery to Stripe checkout — in parallel.
Here's the architecture.
The Stack
- OpenAI Codex / Claude Code — autonomous coding agents that scaffold TypeScript MCP servers from templates
-
Smithery CLI —
npx smithery mcp publishdeploys to their marketplace in one command -
npm —
@rumblingb/*namespace for discoverability - Stripe Payment Links — $19/mo Pro, $99/mo Unlimited, no code required
- Hermes (my AI command center) — orchestrates all of this as a cron job
The Pipeline
Idea → Codex builds repo → npm publish → Smithery deploy → Stripe product + price → Payment link live
Each step is a shell command or API call. There's no dashboard. No manual clicking. Agents execute the whole thing.
The Economics
61 products. Each with a Stripe link. Sound like a lot? Most took under 10 minutes to ship.
- Email Verify MCP — validates emails via DNS, zero API cost, pure profit if anyone subscribes
- IP Geo MCP — geolocation from IP, same model, free to run
- VAT Validation MCP — EU VAT number checker, regulatory requirement, built-in demand
These aren't AI wrappers. They're data utilities with real business use cases.
What I Learned
Shipping volume beats perfection. You don't need the perfect product. You need 50 decent ones and 1 that resonates.
Stripe Payment Links are underrated. Everyone builds checkout pages. I paste a link. Done.
MCP is early. The ecosystem has maybe 5,000 servers total. Being one of the first 50 publishers on Smithery gives you distribution leverage.
Agents ship faster than you think. A Codex agent can scaffold, test, and deploy a new MCP server in 3-5 minutes. The bottleneck is ideas, not execution.
The Directory
I put all 61 products in a searchable directory: rumblingb.github.io/mcp-server-directory
Smithery handles discovery: smithery.ai/servers/vishar-rumbling
What's Next
- Auto-generate marketing content for each product (Dev.to articles, social posts)
- A/B test pricing ($19 vs $29 Pro tier)
- Add usage-based billing for high-volume servers
- Build the AgentPassport — governed payment middleware so AI agents can pay each other
If you're building MCP servers, stop overthinking. Ship 10 this week. One of them will stick.
Top comments (2)
This is a fascinating build-in-public example because it shows agents as a production pipeline, not just a coding assistant. Once you have Codex/Claude generating repos, publishing to npm, and deploying to marketplaces, traceability becomes really important. I’d be curious how you track which agent produced which change, which command ran, and where failures happened across the pipeline. That execution history could become just as valuable as the generated code itself.
That’s exactly the part I’m realizing becomes the real infrastructure layer.
Right now Hermes logs every stage transition (repo creation, npm publish, Smithery deploy, Stripe provisioning, etc.) plus agent outputs and shell execution results. The hard lesson was that once agents can ship autonomously, debugging turns into an observability problem more than a coding problem.
I started treating the pipeline more like CI/CD for autonomous workers:
One broken npm publish across 61 parallel jobs can silently cascade if you don’t build traceability in from day one.
I also think there’s a bigger opportunity here: “GitHub Actions for AI agents.” Not just generating code, but creating auditable execution histories for autonomous systems.
The generated code is replaceable. The orchestration + reliability layer probably isn’t.