Agent-native
Software designed to be operated by an AI agent through typed tools, rather than by a human through a dashboard: the interface is the protocol, not the UI.
Agent-operated software exposes structured operations so an assistant can inspect state, request changes, and read outcomes. The server validates what an operation accepts; the assistant and operator still need to choose the right action and interpret its result.
In practice
Ask a connected assistant to inspect a saved group and explain the result before requesting a change.
What to check
Typed tools constrain accepted inputs, but they do not validate business judgment or prove a person approved a call.
Use it in LinkedBoost
- get_lead_group
Inspect saved members, eligibility, and research-column coverage.
- list_agent_activity
Review the workspace's recorded tool calls across MCP, CLI, and HTTP.
Related terms
- Envelope response
- A consistent structured reply shape from every tool call: success flag, data, a one-line summary, notable signals, and durable learnings: so an agent always knows what happened and what deserves attention.
- Model Context Protocol (MCP)
- The open standard (launched November 2024) that lets AI agents call external systems through typed tools: the USB port between models like Claude and the software that does things.