apimesh
single-purpose dev tools that solve one specific pain. each wedge stands alone — its own subdomain, its own demo, no signup.
the problem
started early 2026 as a marketplace of pay-per-call web analysis apis with an autonomous build loop. the marketplace framing didnt land — human devs wanted flat pricing on focused tools, not crypto micropayments on a hundred half-built endpoints, and the autonomous build loop generated more breadth than signal.
pivoted in april 2026 to single-pain wedge products. each wedge is its own focused tool with its own subdomain, its own seo term, its own honest scope. the umbrella brand recedes — the wedges are the products.
how it works
wedges ship standalone on their own subdomains like agentsmd.apimesh.xyz. shared chassis under the hood — caddy routing, bun + hono, single $4/mo hetzner arm server — but each wedge is independently deployable with no blast radius if one breaks.
wedges are picked by pain not by topic. if a developer hits the same wall three times and there isnt a small focused fix already, that's a wedge candidate. if a wedge grows past one specific job, it splits into a new one.
- agentsmd — fan AGENTS.md out to CLAUDE.md, .cursorrules, .clinerules, .windsurfrules. cli + mcp server, no copy-paste tax.
- stripesig — paste a failing stripe (or github / slack / shopify) webhook, get a plain-english reason and the fix in 5 seconds.
- each wedge on its own subdomain, independently deployed, independently brandable.
- shared chassis: caddy + bun + hono + sqlite on one hetzner arm. mcp server published to npm as @mbeato/apimesh-mcp-server.
what shipped
two live wedges as of may 2026 — agentsmd and stripesig — each on its own subdomain with its own demo and seo target. mcp server on npm.
the pivot has a real cost to own: ~90 apis from the original marketplace are being culled by a 14-day-zero-revenue prune script. lesson taken — ship fewer focused tools that earn their pain instead of generating breadth autonomously.
decisions and tradeoffs
the biggest decision was killing the marketplace strategy after six months of building it.
- pivoted from marketplace to single-pain wedges in april 2026. the marketplace framing assumed agents would be the buyers; turned out human devs wanted flat pricing on focused tools.
- turned off the autonomous build loop. it generated breadth without signal — most apis got zero usage. the prune script removes any wedge that goes 14 days with zero revenue.
- each wedge gets its own subdomain and own seo term over a unified marketplace. lets each tool earn its own pain on its own.
- kept x402 and stripe mpp wired up as free upside, not the primary pitch. agent payments are a real ecosystem but not yet a real distribution channel.