Skip to content
DromlikDocs
sv

Add an MCP server

Last updated: 2026-06-30

An MCP server bundles a set of tools, data sources, and services and exposes them to the Dromlik AI Receptionist over the Model Context Protocol. Instead of registering each API endpoint as a separate custom tool, you point the receptionist at one MCP server URL and it picks up every capability that server publishes.

Requirements

  • Dromlik tenant on release v25.2 or later.
  • An MCP server reachable over HTTPS using one of these transports:
    • Streamable HTTP
    • Server-Sent Events (SSE)
  • The MCP server's endpoint URL (must start with https://).
  • If the server needs auth, the static credential to send in a header — for example an API key or a static Bearer token. Dynamic auth flows such as OAuth 2.0 are not supported because tokens can't be refreshed at runtime.

Limitations

  • Up to 20 MCP servers per tenant.
  • Up to 20 static headers per MCP server.
  • Only static header values — no per-call variables.

Steps

  1. 1

    Open the MCP server list

    Sign in to the Dromlik admin console and go to AI > AI Receptionist > MCP Server. Click Add.

  2. 2

    Fill in the basics

    In the Basic section:

    • Name — a label that helps you identify this server later.
    • Server type — pick Streamable HTTP or SSE to match the endpoint.
    • URL — the full https:// endpoint URL.
    • Remark — optional note (purpose, environment, owner).
    • Description — a natural-language description that tells the receptionist when and how to use this server. This is what the model reads when it decides whether to call it, so be specific.
  3. 3

    Set the execution rules

    In the Execution rules section:

    • Disable interruptions during tool execution — callers can't barge in while a tool is running.
    • Play prompt before executing tool — always speaks a short filler (e.g. "Let me check that for you") before calling the tool. When off, the receptionist decides based on recent tool latency.
    • Execution mode:
      • Execute immediately on LLM request — run as soon as the model asks.
      • Execute after voice prompt — run after the filler prompt.
      • Async — fire-and-forget; the conversation continues without waiting for a response. Best for side-effects like sending an email or triggering a workflow.
    • Tool execution sound effect — ambient sound during execution (None, Typing sound, or Music 1–4). Not available in Async mode.
    • Sound effect playback rules — play only after a voice prompt, or always.
  4. 4

    Add headers (optional)

    In the Headers section, add any static headers the server needs — typically for authentication. Examples:

    • Authorization: Bearer <your_static_token>
    • X-API-Key: <your_api_key>
  5. 5

    Save

    Click Save. The server appears in the MCP server list.

Verify the connection

In the server list, the Status column shows a green Connectedindicator once Dromlik has established a secure connection to the remote MCP server. Click the view icon next to the server to see the list of tools it publishes — this is the same list the receptionist will see at call time.

What to do next

Assign the MCP server to an AI Receptionist under its skill configuration, then place a test call to confirm the receptionist can discover and call the server's tools in a real conversation.