anchorConfigure MCP Client Action

The Model Context Protocol (MCP) client capability enables autonomous AI agents to connect directly to third-party tools and services without requiring custom-built API integrations. This streamlines how agents retrieve data and perform actions during live customer conversations.

By setting up an MCP client, your AI agents can:

• Automatically invoke third-party tools during live conversations
• Retrieve real-time data from external systems
• Perform actions without human intervention

Prerequisites

Before you configure an MCP action in the Webex AI Agent Studio, you must have:

• The MCP server URL
• Authentication type and credentials for the third-party service

Note: MCP actions are read-only once created. If you need to change action settings, you must create a new action.

Steps

1. Log in to the Webex AI Agent Studio platform.

2. Choose the AI agent for which you want to configure the MCP action. You can also create a new autonomous AI agent. See the Create an autonomous AI agent section for more information.

3. Navigate to the Actions tab. Click + Add actions.

4. In the Add actions pop-up, under Browse actions, choose Select available.

This displays all the available MCP tools with the provider name and MCP label.

5. Choose the MCP tool you want to add. You can add multiple MCP tools at any time. Click Add.

Your agent can support a maximum number of actions as defined by your organization's configuration.

What to Do Next

Once an MCP tool is added, the system automatically populates the following details from the MCP server definition:

• Action name — MCP tool name
• Description
• Input entities — Required fields and data types

If you need to modify any of these details, you must update the MCP server definition at the registration or provisioning stage. Changes get synchronized automatically to the AI Agent Studio via notifications.

Verify MCP Tool Status

After adding an MCP action, verify its status in the action list:

• Tool name, description, and server details
• Input parameters sent to the tool
• Output returned by the tool
• Execution state (Success or Failure)
• Latency (in milliseconds)
• Server name and ID
• Transaction ID

Limitations

• Group-level access control is not supported. MCP tool access is managed at the organization level only.
• MCP actions are read-only once created in AI Agent Studio. To update an action, create a new one.
• Maximum number of actions per agent is defined by your organization's configuration. Contact your administrator for details.