fleetcheck-mcp
The problem
When a store calls saying the internet is down, the first five minutes are always the same ritual: ping the gateway, check DNS, hit the health endpoint, test the VPN port, look at the TLS cert quietly counting down to expiry. Doing that by hand across a fleet is slow, and it is the part of the job least worth a human’s attention.
What I built
An MCP server that hands those checks to an AI assistant as seven read-only tools — ping_host, dns_lookup, http_health, tcp_port_check, tls_cert_check, traceroute_host and a fleet_sweep that runs them across many hosts at once. Read-only by design: it can tell you what is broken, never change it. Hostnames and ports are validated before anything touches the network; failures come back as structured data, never exceptions; sweeps are capped at 50 hosts and 10 concurrent checks.
The result
Published under MIT with CI green on Python 3.10–3.12 and 107 offline tests against canned ping output. It is the bridge between the two halves of my work — the outage playbook I actually run, packaged so an AI assistant can run it too. Installable from Git; not yet on PyPI.