Tool interruptions

Overview

By default a user can interrupt the agent at any point, including while a tool is executing. For some tools this is undesirable: a payment confirmation or a compliance disclaimer read out after a lookup should be delivered in full.

The interruption_mode field controls this per tool. It is available on webhook tools, client tools, system tools, and MCP servers.

Interruption modes

Value	Dashboard label	Behavior
`allow`	Allow	The user can interrupt the agent at any time. This is the default.
`disable_during_tool`	Disable during execution	Interruptions are suppressed only while the tool runs. The agent response that follows can be interrupted.
`disable_during_tool_and_turn`	Disable during whole turn	Interruptions are suppressed while the tool runs and for the agent response that follows it.

When several tools run in parallel, the strictest mode among them applies for that turn.

Configuration

Update via the dashboard

Navigate to tool configuration

Select the tool you want to configure in the Tools tab.

Set the interruption mode

Open Advanced settings and choose an option under Interruptions.

Save your configuration

Click Save to apply the change.

Update via the CLI

Pull the tool configuration

Bash

elevenlabs tools pull

Edit `tool_configs/<tool-name>.json`

Set interruption_mode on the tool:

JSON

{
  "type": "webhook",
  "name": "confirm_payment",
  "description": "Confirms a pending payment",
  "interruption_mode": "disable_during_tool_and_turn"
}

Push your changes

Bash

elevenlabs tools push

Update via the API

Python

from elevenlabs import ElevenLabs, ToolRequestModel

elevenlabs = ElevenLabs()

elevenlabs.conversational_ai.tools.update(
    tool_id="tool_7101k5zvyjhmfg983brhmhkd98n6",
    request=ToolRequestModel(
        tool_config={
            "type": "webhook",
            "name": "confirm_payment",
            "description": "Confirms a pending payment",
            "interruption_mode": "disable_during_tool_and_turn",
            "api_schema": {
                "url": "https://api.example.com/payments/confirm",
                "method": "POST",
            },
        },
    ),
)

TypeScript

import { ElevenLabsClient } from "@elevenlabs/elevenlabs-js";

const elevenlabs = new ElevenLabsClient();

await elevenlabs.conversationalAi.tools.update("tool_7101k5zvyjhmfg983brhmhkd98n6", {
  toolConfig: {
    type: "webhook",
    name: "confirm_payment",
    description: "Confirms a pending payment",
    interruptionMode: "disable_during_tool_and_turn",
    apiSchema: {
      url: "https://api.example.com/payments/confirm",
      method: "POST",
    },
  },
});

MCP servers

An MCP server carries its own interruption_mode, which applies to every tool it exposes. Individual tools can override it through the server's tool_config_overrides, where the value set on a tool takes precedence over the server default.

Best practices

Use disable_during_tool when the tool call itself must complete, such as a transfer or a keypad sequence, but the user should be able to interrupt the response afterwards.
Use disable_during_tool_and_turn when the response that follows the tool must be heard in full, such as a disclaimer or a confirmation read-back.
Leave the default allow in place elsewhere, since suppressing interruptions makes the conversation feel less responsive.

Search documentation

Help Center