MCP server for shared context caching — AI agents share computed results to reduce token cost and latency
