MCP & Tooling
Agent Harness
Also: agent runtime, agent scaffold
The program around a model that gives it tools, runs its tool calls, manages context and enforces permissions.
In more detail
Claude Code, Codex CLI and Cursor are harnesses wrapped around a model. Most of an agent's real-world behavior, including what it may run without asking, is set by the harness and its configuration, not the model alone. That is why the same model can be safe in one harness and risky in another.