The Agent SDK query builder accepts a permissionPrompts option of host or none, passed through to the CLI.
In --print/headless mode, a new --permission-prompts <target> CLI option lets you choose who answers permission prompts: host (the default, using the SDK host or --permission-prompt-tool) or none. With none, the SDK does not consult the host; anything that would need a prompt is auto-denied locally, hostAnswersElicitations is disabled, and the SDK logs --permission-prompts none: permission prompts are answered with a local deny; the SDK host is not consulted. Other permission-mode behavior is unaffected. The Agent SDK / CLI query builder exposes the same choice as a permissionPrompts option ("host"/"none"), passed through as --permission-prompts to the CLI. The SDK-level none option is separately marked unsupported for cloud/remote sessions today, logging --permission-prompts none (turning permission prompts off is not supported for a cloud session yet; they reach the host over stdio).
When a tool requiring permission is hit under --permission-prompts=none (or any session with no approval surface), it is now auto-denied with an explicit message telling the model the action was not performed, not to retry, and that all further approval-requiring actions will be denied for the rest of the session.
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
If you set `permissionPromptToolName`, the first turn also waits for that tool's server in every case, up to `MCP_TIMEOUT`.agent-sdk/mcp see the edit
Anthropic's documentation has since written up permissionPromptToolName, on Connect to external tools with MCP.
Added --permission-prompts none for unattended headless hosts: anything that would prompt is denied automatically while the active…