Where it is raised
- Keyless MCP endpoint, the per-address budget
What triggers it
The keyless endpoint allows 60 requests a minute per client address. A request carrying several tool calls spends one unit per extra call, so batching does not get around the limit, and a refusal can come before the body is read or after it is inspected.
Example response
POST https://mcp.realuptime.io/public
HTTP 429
retry-after: <seconds>
{
"error": "Rate limit exceeded on the keyless MCP endpoint. Wait and retry, or use an API key for a larger per-key budget.",
"code": "RU-6004",
"docs": "https://realuptime.io/kb/errors/ru-6004",
"retryAfterSeconds": "<seconds>"
}Common causes
- An agent polling an outage tool in a loop instead of asking once.
- Many callers behind one shared egress address, which share a single per-IP budget.
- Protocol overhead: one assistant question costs several requests (initialize, tools/list, then the tool call).
How to fix it
- Honor the Retry-After header before retrying.
- Ask once and reuse the answer: readings only change once per probe cycle (10 minutes).
- For sustained or shared-egress use, switch to the keyed MCP endpoint with an API key, which has its own larger per-key budget.
- For a single service reading, the cacheable JSON endpoint at /api/v1/outages/<slug> is cheaper than an MCP round trip.