Add an MCP Server

An MCP Server provides a bundled set of tools, services, and data sources via the Model Context Protocol (MCP), enabling the AI receptionist to access an entire suite of capabilities through a single connection. This topic describes how to add an MCP server to Yeastar P-Series Cloud Edition.

Requirements

Table 1.
Server Requirement
PBX Server The firmware version of Yeastar P-Series Cloud Edition is 84.24.0.30 or later.
MCP Server
  • The MCP server must support one of the following transport mechanisms:
    • Streamable HTTP
    • Server-Sent Events (SSE)
  • The following information must be obtained from the MCP server:
    • The endpoint URL of the MCP server.
    • If the MCP server requires authentication, obtain the static credentials (such as API key, Basic Auth username and password, or static Bearer Token).
      Note: Dynamic authentication (e.g., OAuth 2.0) is not supported, as it requires dynamic token acquisition and cannot be pre-configured.
      Tip: You can pass the authentication credentials via the header parameters in the configuration. For example, Authorization: Bearer <your_static_token>, X-API-Key: <your_api_key>, etc.

Limitations

The PBX supports a maximum of 20 MCP servers.

Procedure

  1. Log in to PBX web portal, go to AI > AI Receptionist > MCP Server.
  2. Click Add.
  3. In the Basic section, configure the basic information.

    Table 2.
    Setting Description
    Name Specify a name to help you identify the MCP server.
    Server Type Select the transport mechanism that matches the MCP server's endpoint URL.

    For example, if the URL points to a Streamable HTTP endpoint, select Streamable HTTP.
    URL Enter the complete endpoint URL of the MCP server.
    Note: The URL must start with https://.
    Remark Optional. Enter a note for quick reference, such as purpose, environment, or owner.
    Description Enter a natural language description to instruct the AI receptionist on when and how to use this MCP server.
  4. In the Execution Rules section, configure the execution rules for the MCP tools.

    Table 3.
    Setting Description
    Disable Interruptions During Tool Execution When selected, callers can not interrupt the AI receptionist while the MCP tool runs.
    Play Prompt Before Executing Tool When selected, the AI receptionist always speaks a prompt (e.g., "Let me check that for you") before running the MCP tool.

    When unselected, the AI receptionist decides whether to speak a prompt automatically based on the recent execution latency of the MCP tool. For example, skipping it for fast responses and speaking first for slow ones.
    Execution Mode Determines the execution behavior of the MCP tool.
    • Execute Immediately on LLM Request: The tool runs as soon as the AI receptionist requests it.
    • Execute After Voice Prompt: The tool runs after the AI receptionist plays a voice prompt.
    • Async: The tool runs in the background without pausing the conversation. The AI receptionist does not wait for the response and continues speaking immediately.
      Note: This mode is best suited for operations that don't require a response, such as sending an email or triggering an external workflow.
    Tool Execution Sound Effect Specify the ambient sound during MCP tool execution.
    Note: This setting is not available when Execution Mode is Async.
    • None: No sound will be played.
    • Typing Sound: Play a keyboard typing effect.
    • Music 1 - Music 4: Play one of the built-in ambient sound.
    Sound Effect Playback Rules
    Specify when to play the ambient sound.
    Note: This setting is not available when Execution Mode is Async or Tool Execution Sound Effect is None.
    • Play Only After Voice Prompt: Plays only when the AI receptionist speaks a prompt before executing the tool.
    • Always Play: Plays during every tool execution, regardless of whether the AI receptionist speaks a prompt first.
  5. Optional: In the Headers section, add and specify headers to be included in the request.
    Note:
    • Only supports static header values. If authentication needs to be passed in headers, use static credentials (e.g., Authorization: Bearer <Your_API_KEY>).
    • Supports up to 20 headers.
    Table 4.
    Setting Description
    Parameter Header name.
    Values Fixed header value.
    Note: Dynamic values are not supported.
  6. Click Save.

Result

The MCP server has been added successfully and appears in the server list.
  • You can verify the connection status in the Status column. A (Connected) status indicates that a secure connection to the remote MCP server has been established.

  • You can click beside the MCP server to view the available tools provided by the server.

What to do next

Assign the MCP server to an AI receptionist and test it in a conversation.