Compositions

MCP + Skills

The server supplies a narrow capability; the skill defines when and how to use it. Keep authorization and input validation on the server.

MCP + Subagents

Give each specialist only the servers and tools required for its role. A research worker should not inherit production mutation tools.

MCP + Hooks

Use client hooks for fast observation or local policy where supported. Record and enforce the external action again at the server because not every surface traverses the same hook path.

MCP + Memory

Memory may store stable identifiers or learned routing context. Re-query the authoritative system for volatile state; do not persist secrets or treat cached output as current truth.

MCP + Approval Service

For consequential writes, the agentAgentA worker you delegate to: brief it, and it takes steps on its own. A chatbot answers; an agent acts.An LLM that runs tools in a loop toward a goal — it acts, checks the result, decides the next step, and repeats until done.Full definitionClose proposes a typed action. An independent service evaluates policy and human approval, then performs or authorizes the external mutation.

MCP + Evidence Ledger

Link the agent request, policy decision, server invocation, downstream result, and verification under one causal identifier. This turns tool useTool useLetting the model pick up the phone. On its own it can only write text; a tool lets it do something in the real world.Also called "function calling": the model requests an action from outside software — search the web, read a file, hit an API — and the result comes back into its context.Full definitionClose into an auditable delivery event.