Model Context Protocol (MCP) inbound allows you to start automation in your bot processes or automations from third-party AI Assistants. This is done securely through a multi-tenant, user-aware MCP gateway to start Automation Anywhere automations.

With MCP inbound connectivity, Automation Anywhere can receive a request from third-party AI Agents and third-party AI Assistants. Using MCP, an AI Agent can:

  • Get data from third-party systems.
  • Make decisions using company data.
  • Run and discover automations from the Control Room.

Prerequisites

Make sure you have set up the following:

  • Control Room repository permissions.
  • An MCP client: For example, Microsoft Copilot, ENTKB (Enterprise KB), or another MCP-compatible agent that can register an external MCP server and use tools. Each MCP client might support different MCP features, allowing for varying levels of integration with MCP servers. For a comprehensive list of supported MCP clients and latest features, click Feature support matrix.

Procedure

  1. Sign in to the Control Room.
  2. Go to AI > Agent connections. From the Agent connections page, you can view the list of agent connections and tools.

    Agent add

  3. To add a new agent connection, you can either:
    • At the top right of the page, click + Add > Inbound tool.
    • At the right of the page (above the table), click the Create icon.
  4. From the Add inbound tool page, click Browse to search for the automation from a list of folders and files.
  5. Click to select the automation you want and then click Choose.
  6. After you select the automation:
    • If you have the Enterprise license with Process Reasoning Engine (PRE) (see Motor de razonamiento de procesos y generative AI) enabled, then the Tool name and Description fields are automatically populated with content from PRE. Our MCP inbound interactions are powered by our PRE. The PRE has a deep understanding of all the Automation Anywhere automations, processes, API Tasks, and AI Agents. It knows what they are and what they can do. When a user requests to use an MCP inbound tool from a third-party AI assistant, our PRE understands the user's intent and the context of the request. Then, it matches this with the unique automation in the Automation Anywhere repository.
    • If you have the Base license, then the Tool name and Description fields are automatically fetched from the repository.
    • However, if there is no description in the repository, then it will be blank. Both fields are required.
      • Tool name (Required): Rename to something AI-friendly and make it easier for the third-party AI Agent to understand. The name must be unique across all Agent connections. It must be between 1-128 characters and only contain letters (A-Z, a-z), digits (0-9), underscore (_), hyphen (-), or dot (.). No spaces or special characters are allowed.
      • Description (Required): Keep it clear and action-oriented.
  7. Click Add.
  8. Your new inbound tool now appears in the Agent connections table with a connection status of Enabled.
    Note:

    If an automation is updated or deleted, then you must update or delete the agent connection manually.

    Supported Input and Output Variable Data Types
    • From MCP Inbound for Task bots and API Tasks:
      • Input: String, Number, Boolean, List, Dictionary, DateTime, and Record
      • Output: String, Number, Boolean, List, Dictionary, DateTime, and Record
    • From MCP Inbound for processes:
      • Input: String, Number, Boolean, List, Dictionary, and DateTime
      • Output: No data types are supported for process output.