ClaudeAgent wraps the Claude Agent SDK so Claude Code can be served through AgentOS.
Tool execution is handled by the SDK. You configure which built-in tools are allowed and which permission mode to use.
Install
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
name | str | None | Display name for the agent. |
id | str | None | Unique identifier. Auto-generated from name if unset. |
system_prompt | str | None | Optional system prompt. |
model | str | None | Model id (e.g. "claude-sonnet-4-6"). Defaults to the SDK default. |
allowed_tools | List[str] | None | Built-in tools the agent can call (e.g. ["Read", "Bash", "WebSearch"]). |
disallowed_tools | List[str] | None | Tools to block. |
permission_mode | str | None | One of "default", "acceptEdits", "plan", "bypassPermissions". |
max_turns | int | None | Maximum number of turns per run. |
max_budget_usd | float | None | Hard cost cap per run. |
cwd | str | None | Working directory the agent runs in. |
mcp_servers | Dict[str, Any] | None | MCP server configurations for custom tools. |
options_kwargs | Dict[str, Any] | {} | Extra kwargs forwarded to ClaudeAgentOptions. |
db | BaseDb | None | Database for session persistence. |
Examples
AgentOS deployment
Serve a Claude Code agent through AgentOS.
Standalone usage
Call the agent directly with
.run() and .print_response().Custom MCP tools
Extend Claude Code with your own MCP servers.
Sessions
Resume conversations across runs with
session_id.