Part of #5677

Spec link: https://modelcontextprotocol.io/specification/draft/changelog
Comprised of: Minor change 3 (deterministic tools/list ordering) + Minor change 5 (CacheableResult — SEP-2549)

What the spec says

• Results returned by tools/list, prompts/list, resources/list, resources/read, and resources/templates/list must include ttlMs and cacheScope via a new CacheableResult interface.
• ttlMs is a freshness hint in milliseconds allowing clients to cache responses and reduce polling.
• cacheScope ("public" or "private") controls whether shared intermediaries may cache the response.
• Both fields complement — not replace — the existing listChanged notifications.
• Servers SHOULD return tools from tools/list in a deterministic order to enable client-side caching and improve LLM prompt cache hit rates.

Gateway work

• Add ttlMs/cacheScope to all applicable gateway responses with sensible, config-overridable per-endpoint defaults.
• As a shared intermediary, honor upstream cacheScope: never place "private" results in shared cache tiers.
• Aggregated responses: ttlMs = minimum across contributing upstreams; cacheScope = "private" if any contributor is private.
• Guarantee stable, deterministic ordering in aggregated tool/prompt/resource lists (relevant to gateway prompt-cache friendliness too).

Acceptance criteria

• All list/read responses carry valid CacheableResult fields.
• "private" upstream results never enter shared cache tiers.
• Repeated tools/list calls return byte-identical ordering absent registry changes.