Two milestones worth separating from the noise. The API got an MCP server - a typed tool interface an AI agent can drive - so "assistant that can actually do things in the platform" became possible instead of theoretical. And we built status.marsscuba.com from nothing.
The monitors do not just ping a URL - they actually drive the product: sign a waiver, submit a medical, reset a password, run the PDF pipeline - and scream when any link in the chain breaks. I was tired of finding out things were broken from customers.
Shipped
- /v1/mcp - the platform's first AI-agent interface
- Recognition Program engine: schema, reconciliation, CRUD, triggers on certification-ID writes
- Full service-estimate flow: send estimate → public customer approval (waiver-gated) → QA phase → bulk mark-paid, plus merging multiple work orders into one
- Status app from scratch: DB-queue architecture, local scheduling, log ingestion, and ~20 end-to-end monitors (waivers, medicals, password reset, PDF gen, TLS/DNS, the inbound-SMS signature gate)
- Analytics hardening with PII masking, because "observability" should not mean "leak customer data"