outage.me runs an MCP server, so Claude and other AI apps can ask it what is down, how reliable a provider has been and which exploited vulnerabilities to patch first. The live facts need no account. Signing in adds more tools.
https://outage.me/mcplist_active_incidents, provider_reliability, exploited_vulnerabilities, deploy_check
Live facts only. No model calls.
https://outage.me/mcp/proanswer, provider_history
answer: 10 questions a day. provider_history: last 7 days.
https://outage.me/mcp/probackbone_watch, vendor_scorecard, blast_radius
answer: 100 questions a day. provider_history: 90 days.
Each plan includes the tools of the plans above it.
claude mcp add --transport http outage-me https://outage.me/mcp/pro
Prefer REST? The same facts are open JSON endpoints: see llms.txt and openapi.json.
deploy_check (MCP, no sign-in) and GET /api/deploy-check?provider=®ion=
answer proceed: true, false or "unknown", from the
5-minute status poll, with the open incidents and data_age_minutes. A stale poll, a
failed fetch or an untracked provider is "unknown", never true. A GitHub
Actions step that holds the deploy unless the answer is true:
- name: Hold the deploy during a provider incident (outage.me)
run: |
v=$(curl -fsS "https://outage.me/api/deploy-check?provider=gcp®ion=us-central1" | jq -r .proceed)
echo "outage.me deploy_check: $v"
[ "$v" = "true" ]To hold only on a known incident, end with [ "$v" != "false" ].
Status pages lag real failures: use this to pause a deploy, not to trigger failover.