Skip to content

Agentic AI (MCP)#

conceptstub

Agents that plan, call tools through MCP servers, and chain many model calls to complete one task.

Planned content

  • Architecture: gateway, agent runtime, MCP servers, model endpoints
  • Inference profile: many short chained calls, long shared prefixes
  • KV-cache reuse, prefix-aware routing, and inference gateways
  • Autoscaling and per-tenant rate limits for long-lived sessions
  • Securing tool calls: identity, authorization, audit