conventions

UptimeRobot conventions

Apply these rules whenever you use any tool from the uptimerobot MCP server (https://mcp.uptimerobot.com/mcp).

Monitor types

• Always pass type as a string enum — one of HTTP, KEYWORD, PING, PORT, HEARTBEAT, DNS, API, UDP. Do not pass numeric type codes; the MCP layer rejects them.
• HEARTBEAT monitors must not include a url on creation. The server generates https://heartbeat.uptimerobot.com/m<id>-<hash> and returns it. Show that URL to the user after creation so they can wire up the heartbeat sender.
• PING, PORT, UDP, DNS: url is a hostname or IP, never a full URL with scheme.
• HTTP, KEYWORD, API: url must be a full URL with scheme (https://...).

Discovery before writes

• Before calling update-monitor, update-monitor-status, or referencing a monitor by ID, call list-monitors (optionally with search) to confirm the ID and current state.
• Before referencing an alert contact in assignedAlertContacts, call list-integrations and use an id from the response.

Verification after writes

• After create-monitor or update-monitor, call get-monitor-details to confirm the stored state before telling the user the change is live. Read-after-write replication can lag a few seconds.

Intervals and plan limits

• Interval is in seconds. Minimum is 30s and depends on the plan. If the server returns -28002 subscription_limit_exceeded, do not retry — the plan cannot support that interval or monitor type. Tell the user.
• Do not retry -28001 monitor_limit_exceeded. Recommend deleting unused monitors or upgrading the plan.
• Plans are Free, Solo, Team, Enterprise. Never reference a "Pro" or "Business" plan — neither exists.

Time ranges

• timeRange accepts "24h", "7d", "30d" (range 1h–90d), or an ISO 8601 interval "<start>/<end>" (start before end, 1h–90d). Default is "7d".
• For anything older than 90 days, inform the user that UptimeRobot's MCP endpoints cap at 90 days of history.

Pagination

• Paginated tools (list-monitors, list-incidents, list-integrations) return an instructions field with the exact next-call JSON. Follow it verbatim; cursor shape is opaque.

Errors

• -28001 monitor limit, -28002 subscription limit, -29001 invalid params (incl. blacklisted URLs), -30001 / -30003 not found, -31002 access denied (authorized account lacks write permission).
• -31001 user_not_found means the OAuth grant is missing/expired/revoked — do not surface it raw; invoke the uptimerobot:setup skill to re-authenticate.
• HTTP 429 is the only code worth retrying (exponential backoff + jitter). Everything else is a client error — surface it to the user.

Output style

• Do not echo API keys, incident log contents, or traceroute hops unless the user explicitly asks.
• When summarising many monitors, group by state (DOWN first, then PAUSED, then UP) and show counts before detail.
• When quoting uptime percentages, always cite the time range.