How can we help?

AI Agent Platform: Build - Skill (MCP)
Published • Last Updated

The Model Context Protocol (MCP) integration allows Talkdesk users to expand the scope of AI Agents using externally hosted tools, enabling integration with custom logic, APIs, and data sources outside of Talkdesk. By leveraging MCP, customers can reuse existing tools or expose their own capabilities to AI Agents in a standardized and scalable way.

MCP provides a uniform mechanism for tool definition and remote execution, allowing AI Agents to invoke preconfigured tools through a customer-hosted MCP Server.

 

Table of Contents

 

What is MCP?

Model Context Protocol (MCP) is an open protocol designed to simplify how AI agents discover and invoke tools. MCP abstracts tool management behind an MCP Server, which exposes tools in a standardized format.

An MCP Server is responsible for:

  • Providing the available tools, including their descriptions and input parameters.
  • Executing tool calls on behalf of the AI Agent and returning structured responses.

In the AI Agent Platform, Talkdesk acts as the MCP Client, communicating with customer-hosted MCP Servers to retrieve tool definitions and execute tools at runtime.

 

Benefits of MCP in the AI Agent Platform

Using MCP brings several advantages when building AI Agents:

  • Remote tool support: Tools are hosted externally, allowing customers to expose custom logic, internal systems, or publicly-available MCP tools without rebuilding them inside Talkdesk.
  • Uniform tool integration: MCP provides a consistent interface for tool discovery and execution, reducing the effort required to add new tools to AI Agents.
  • Improved scalability: Since tools run remotely, customers can scale complex logic without being constrained by the AI Agent Platform runtime.
  • Bring your own tools: Customers can integrate existing systems or proprietary tools without migrating data or logic into Talkdesk.
     

How It Works

Add and Configure an MCP Server

Before AI Agents can use MCP tools, an MCP Server must be added and configured.

  1. Navigate to AI Agent Platform.
  2. From the left side menu, select MCP Servers.

  3. Click + New MCP Server button to open the configuration screen.

     

MCP Server Type

Select the MCP Server type you want to configure. The currently-supported options are:

  • UiPath MCP Server
  • Custom MCP Server

Required Configuration Fields

For both server types, provide the following information:

  • Name: A friendly name to identify the MCP Server.
  • Description: A short explanation of what the server and its tools provide.
  • Status: Enable or disable the MCP Server.
  • URL: The endpoint where the MCP Server is hosted.
    • Transport type supported: Streamable HTTP
  • Authentication Method:
    • Bearer Token
    • OAuth2

Depending on the selected authentication method, please provide:

  • Bearer Token: Token
  • OAuth2:
    • Client ID
    • Client Secret
    • Token URL
    • Scope

You can validate the configuration by clicking the Test connection button.

Once the MCP Server is successfully connected, it will become available for use in AI Agent Orchestrations.

 

Add MCP Tools as AI Agent Skills

After configuring an MCP Server, its tools can be added to AI Agents as Skills.

  1. In the AI Agent Platform, open an existing AI Agent Orchestration or create a new one.
  2. For guidance on creating an AI Agent Orchestration, please refer to the AI Agent Platform documentation.
  3. Within the AI Agent, click + Add Skill.

You can choose between three skill types:

  • Workflow: Used to orchestrate automations using Talkdesk components such as Execute Action, Function, and Split Flow. Workflow skills are executed by a separate runtime outside the AI Agent's reasoning context, making them ideal for complex business logic, loops, retries, and for securely executing JavaScript logic or API credentials that should not be exposed to the AI Agent.
    For more information, see AI Agent Platform: Build - Skill (Workflow).
  • Integration: Used to connect AI Agents to external systems through API calls defined in the Talkdesk Integrations application. Integration skills are best suited to well-defined, synchronous operations, such as retrieving or updating records in third-party systems, where endpoints, authentication, inputs, and outputs are explicitly configured in advance.
    For more information, see AI Agent Platform: Build - Skill (Integration).
  • MCP: Used to connect AI Agents to tools exposed by a remote Model Context Protocol server. MCP skills allow AI Agents to invoke externally hosted tools, enabling access to custom logic, proprietary systems, or publicly available MCP tools without implementing them directly in Talkdesk. Each MCP skill represents a single remote tool and runs outside of the Talkdesk platform, with execution fully managed by the customer-hosted MCP Server.

Select MCP and then:

  • Choose the previously configured MCP Server.
  • Select one MCP tool from the list returned by the server. Each MCP tool maps to a single AI Agent skill.
  • Provide a Tool Name and Description to help the AI Agent understand when to use the tool.
  • Configure any required input variables that will be passed to the tool at runtime.

You can add multiple MCP skills to the same AI Agent, either from the same MCP Server or from different servers.

After clicking Save, the MCP skill becomes immediately available in the orchestration and is visually identified with a distinct skill icon.

Test and Publish

Once the MCP skill is configured:

  • Use the Test option at the top of the orchestration screen to simulate interactions and validate tool behavior.
  • When the results meet expectations, Publish the orchestration to make it available for production use.

 

Use Case

A common example is an online shopping AI Agent that uses MCP tools to manage orders. An MCP Server exposes tools such as:

  • Add item to cart
  • Remove item from cart
  • Get prices
  • Read cart
  • Place order

In this setup, the user explicitly selects which tools are connected to the AI Agents and defines when they should be invoked. The AI Agent calls the configured MCP tools with the approved inputs and receives structured responses from the MCP Server, enabling order management without embedding the business logic directly in Talkdesk.

 

Limitations

While MCP enables powerful integrations, there are some considerations to take into account:

  • MCP tools count toward the AI Agent tool limit.
  • Tool definitions are managed externally, so changes on the MCP Server side are not versioned within the AI Agent Platform.
  • Authentication and security must be carefully managed, since MCP Servers are hosted remotely.
  • Only the Streamable HTTP transport type is currently supported.
All Articles ""
Please sign in to submit a request.