A call over the rate or the concurrency limit gets
429 rate_limited. A call past the deadline gets 503 busy. Both carry Retry-After in seconds. Set your client’s timeout to at least 15 seconds.
The limits are the same over MCP, and every MCP request counts, tools/list included. Over MCP, a 429 is an HTTP status; a call past the deadline comes back as a tool error that says to retry.
Every REST reply that was served carries X-Credits-Cost and X-Credits-Remaining. What each datapoint costs is on Pricing.