1. Add the keyless server to your client
Point any MCP client at https://mcp.realuptime.io/public with no credentials at all. In Claude Code that is: claude mcp add --transport http realuptime-outages https://mcp.realuptime.io/public. In a JSON config, add an mcpServers entry with just that url and no headers.
2. Ask the question
Ask your assistant something like "is Stripe down right now?". It calls is_service_down and gets back our own probe measurements from up to four regions, with a timestamp and a URL to cite. The other three tools cover per-region detail, how many people are reporting problems, and past outages we detected.
3. Read the answer correctly
A "blocked" reading means our probe was refused, which is evidence about our access and never a confirmed outage. A no-data answer is never a healthy answer. A service we do not track answers "not covered", not a guess. These are measurements, not user reports, and the two are never merged into one number.
4. Know the limits, and when to use a key
The keyless endpoint allows 60 requests per minute per IP address, and one question costs about four requests. Over budget answers RU-6004. It serves the four outage tools only: anything that reads or changes your own monitors needs an API key on https://mcp.realuptime.io/mcp, and answers RU-1006 without one.
Go deeper
The full reference lives in the docs: MCP server documentation. Error codes named above are each explained in the error-code reference.