Statuspage MCP server

@pipeworx/statuspage

Service status / uptime MCP — "is X down right now?" answered from vendors' own Atlassian Statuspage feeds. Keyless.

Part of Pipeworx — an MCP gateway connecting AI agents to 1476+ live data sources.

Atlassian Statuspage is a de-facto standard: hundreds of vendors expose the same JSON contract at https:///api/v2/status.json. This pack reads that contract, so it closes a class of question rather than one vendor.

Tools

one vendor: overall indicator, the vendor's own status line, every degraded component, open incidents with the latest update text, and upcoming maintenance where published.

• statuspagecheck(vendor | statushost, include_components?) — current status for

— incident history: title, lifecycle status, impact, start/resolve times, duration, affected components, latest update.

• statuspageincidents(vendor | statushost, unresolvedonly?, limit?, includeall_updates?)

pack can answer for, with each one's status host, so a name can be confirmed rather than guessed.

• statuspagelistvendors(category?, query?, include_elsewhere?) — the vendors this

compact up/down table plus a roll-up. Hosts are de-duplicated, concurrency capped at 5.

• statuspagemulticheck(vendors[]) — up to 20 vendors in one call, returned as a

Coverage

Coverage is the curated vendor map plus any caller-supplied host.

The curated map holds 190 vendors across AI/model providers, cloud and hosting, databases and observability, developer tools, payments and fintech, communications, and business SaaS.

Every host in the map was verified, not guessed. 206 candidate hosts were probed live; each survivor had to return a parseable Statuspage v2 payload on /api/v2/status.json, /api/v2/summary.json and /api/v2/incidents.json. Candidates that failed were dropped rather than shipped on a hunch — status hostnames are genuinely unguessable (status.anthropic.com redirects to status.claude.com; status.cloudflare.com does not resolve at all, the real host is www.cloudflarestatus.com).

The second wave (67 vendors, taking the map from 123 to 190) came from a coverage audit against a competing tracker: the bottleneck was never the host pattern, it was knowing which vendor NAMES to try. Given the names, 68 of 78 resolved on the first or second guess of the ordinary pattern, and 67 survived the three-endpoint check (kong was dropped — status.konghq.com answers 418 to a non-browser client).

For anything outside the map, pass statushost: "status.somevendor.com". Every response says which of the two it used via the source field (curatedmap | callersuppliedhost).

Vendors that publish status elsewhere

Roughly 43 well-known companies were probed and found to publish status in some other format — AWS, Google Cloud, Azure, Slack, Stripe, GitLab, Heroku, PagerDuty, Notion, Databricks and others. Asking for one of those returns reason: 'vendornotstatuspage' together with the URL a human should open, rather than nothing. statuspagelistvendors({include_elsewhere: true}) lists them.

Statuspage "lite" hosts

17 hosts in the map serve status.json / summary.json / incidents.json but omit /api/v2/incidents/unresolved.json (OpenAI is one). The pack falls back to filtering the incident feed for open incidents, and reports which path it used in incidents_source.

Result semantics

with operational: true and an empty open_incidents array.

• "All Systems Operational" is a positive result, returned as found: true

same. A fetch failure returns found: false, reason: 'statuspage_unreachable'.

• **A vendor with no incidents and a status page we could not read never look the

near matches and a pointer to statuspagelistvendors — never a guessed host, never a silently substituted vendor.

• An unrecognised vendor returns found: false, reason: 'vendornotcovered' with

Data source

Shortened. The full README is on GitHub.