Skip to main content

​
What is the Claude Agent SDK?

The Claude Agent SDK lets you build your own agents in Python or TypeScript. It runs Claude Code underneath, so an agent can read files, run commands, and edit code. It was previously called the Claude Code SDK. These agents usually run headless, on a server or a VM. Unbound uses the same hooks as Claude Code, so your tool policies apply to every tool call the agent makes.

​
Prerequisites

Before integration, ensure you have:

​
Get your API key

  1. Open Configure in Unbound.
  2. Under API access, copy the API key. If the card is not shown, your key is the value after --api-key in the install command.

​
Setup with Unbound CLI

After setup, start a new agent session.
Setup adds the hooks to ~/.claude/settings.json, which the SDK reads only when settingSources includes "user". That is the default. If your agent leaves out "user", run setup with sudo and an admin API key instead:
To remove the Unbound configuration:
If you installed with sudo, remove it with sudo:

​
Verification

  1. Run unbound doctor. Claude Code should be reported healthy.
  2. Run your agent. Its terminal commands and MCP tool calls show up under Analytics → Agentic Use.

Unbound CLI

Install the CLI to set up and manage tools

Tool Policies

Configure security guardrails for AI tools