Update: 84 → 160 platforms, 920 → 1,532 endpoints in 19 days (and a sync bug that almost shipped)
Quick follow-up to my post from three weeks ago (https://www.indiehackers.com/post/why-i-built-crawlora-after-9-years-of-rebuilding-the-same-crawler-plumbing-SSWOv5Z7WbtfylapCKk3). I said "84+ platforms, 920+ endpoints" then. Today it's 160 platforms, 1,532 endpoints, ~1,400 MCP tools. That's coverage growth, not revenue — just new platforms integrated faster than I expected.
The part worth sharing: growing that fast broke my release pipeline before it broke anything customer-facing.
Every new platform has to propagate from one internal API contract into several SDKs, a hosted MCP server, and public docs. The contract drifted 103 commits between two routine syncs, and when I republished, one about-page README read its number from a stale cached copy sitting in a shared working directory — everything customer-facing got the right count, but that one page quietly didn't. Caught it by eyeballing two numbers that should've matched and didn't.
Fix: never regenerate from a possibly-stale checkout, always a clean pull from origin, and diff every auto-computed number against a known-good source before trusting it.
Anyone else running a multi-surface API (REST + SDKs + agent tools, all generated from one source) — how are you keeping those in sync without silent drift? Rolling your own, or is there tooling for this?
The sync failure is more interesting than the endpoint growth. Once one source feeds REST, SDKs, MCP tools, and docs, consistency becomes part of the product rather than just an engineering concern.