Status
The six services mcprush runs, ninety days of history, and every incident written up by whoever fixed it. A server you installed runs on its publisher's infrastructure — its own record is on its listing.
Platform components
Six services sit between a buyer and a publisher. Each bar is one day, newest on the right, coloured by the worst state recorded that day. Every bar that is not green has a written entry further down — nothing on this page is unexplained.
Every tool call from a client passes through it: the key is checked, the call is routed to the publisher’s server, the response is returned and the call is counted against the plan.
Listing pages, search, the filters, compare and the stacks. Read-only, and the part of the marketplace most people ever see.
Counts calls against the right plan, assembles the Stripe invoice and charges the card on file. We are the merchant of record, so a wrong number here is ours to refund. A Stripe outage shows here as well: nothing can be charged while it lasts, and nothing is charged twice afterwards.
Pulls every release into a sandbox, reads the declared tool surface and produces the security grade printed on the listing.
Assembles the weekly Friday batch and transfers each publisher’s share to their Stripe connected account. Publishers keep 88% of the sale; Stripe pays it out to their bank from there, so a red light here means the batch, not the bank.
What the CLI talks to when it resolves and installs, and what an account key calls to manage its own installs. It serves your account, not the catalogue: nothing here lists or exports the listings.
Uptime subtracts the minutes a component spent returning a wrong answer or no answer, multiplied by the share of traffic that was affected — a region carrying 22% of calls contributes 22% of its own downtime. Time spent slow but correct is not subtracted; it is the amber bar and nothing more. Over 90 days one full minute of total unavailability costs 0.00077%.
Incident history
Written after the fact by the people who fixed it, in the order they happened. What broke, what a reader would have seen from the outside, and what is different now. If a fix was only a revert, it says so.
No incident in the last 90 days. Everything older has rolled out of the window rather than been removed.
Tell me when something breaks
An email when a component leaves operational, another when it returns, and the write-up when it is published.
The address is used for these notifications and nothing else.
Our uptime is not the publisher’s
A server you install almost always runs on its publisher’s own infrastructure. Our gateway forwards the call and returns whatever comes back, so every component here can be operational while the one server you use is not. Each listing publishes its own 30-day success rate and p95 — that is the record for the server itself.
Each component's history is on this page. There is no feed to subscribe to yet — when there is one, its address will be here.