What is an agent harness?
What is it?
A harness is the environment that runs an agent: its instructions file (e.g. CLAUDE.md), the tools it may use, what is denied, hooks and connected MCP servers.
When to use it
- Share a team configuration for Claude Code or another assistant.
- Constrain an agent: allowed commands, protected files.
Page fields
| Harness | Claude Code, Claude Agent SDK, Cursor… |
|---|---|
| Instructions | The instructions file the agent reads. |
| Allowed / denied | Permissions, one rule per line. |
| Hooks | Automatically triggered actions (JSON). |
| MCP servers | Catalog servers to connect. |
How to use it
“Use” tab: copy CLAUDE.md and .claude/settings.json into your project.