Retell AI Server

A fully-featured, official-grade Model Context Protocol (MCP) server that provides complete programmatic access to the Retell AI platform.

This MCP server achieves 100% parity with the Retell Node.js SDK, exposing 60 specific tools allowing LLM-powered assistants (like Claude Desktop, Cursor, and Codex) to autonomously build, configure, and operate end-to-end voice and text agents.

---

Capabilities

With this server, your AI assistant can autonomously:

• Design Agents: Create and configure LLM response engines, Conversation Flows, and Chat Agents.
• Provide Memory: Create Knowledge Bases, upload content, and link them to agents for RAG.
• Manage Telephony: Purchase, import, and configure U.S. or international phone numbers.
• Trigger outbound calls & test: Initiate Web calls, Phone calls, SMS chats, and batch tests.
• Monitor: Retrieve transcripts, real-time metrics, live call concurrency, and latency scores.

Everything available in the official SDK is available as an MCP Tool.

---

Installation

Prerequisites

• Node.js 18+
• A Retell AI API Key (from the Retell Dashboard)

Setup

Clone the repository and install dependencies:

git clone https://github.com/sunnysingh100/retell-mcp-server.git
cd retell-mcp-server
npm install
npm run build

---

Configuration

To use this with an MCP client (such as Claude Desktop or Codex), you must configure it via your client's settings.

Claude Desktop

Edit your claude_desktop_config.json:

{
  "mcpServers": {
    "retell-ai": {
      "command": "node",
      "args": [
        "/absolute/path/to/retell-mcp-server/dist/index.js"
      ],
      "env": {
        "RETELL_API_KEY": "your_retell_api_key_here"
      }
    }
  }
}

Generic Client

Just run the built script with the environment variable set. The server operates over stdio via the standard MCP JSON-RPC protocol.

RETELL_API_KEY="your_api_key" node dist/index.js

---

Available Tools (60 Total)

The server supports all 14 domain resources of the Retell SDK.

🎙 Voices

• retell_list_voices: List available voices across providers (ElevenLabs, OpenAI, Deepgram, etc).

🧠 LLMs & Knowledge Bases

• retell_list_llms / retell_get_llm / retell_create_llm / retell_update_llm / retell_delete_llm
• retell_list_knowledge_bases / retell_get_knowledge_base / retell_create_knowledge_base / retell_delete_knowledge_base
• retell_add_knowledge_base_sources / retell_delete_knowledge_base_source

🤖 Voice Agents

• retell_list_agents / retell_get_agent / retell_create_agent / retell_update_agent / retell_delete_agent
• retell_get_agent_versions / retell_publish_agent

📱 Phone Numbers

• retell_list_phone_numbers / retell_get_phone_number / retell_create_phone_number
• retell_update_phone_number / retell_delete_phone_number / retell_import_phone_number

📞 Voice Calls

• retell_list_calls / retell_get_call / retell_update_call / retell_delete_call
• Initiate: retell_create_phone_call / retell_create_web_call / retell_register_phone_call

💬 Chat Agents & Sessions

• Agents: retell_list_chat_agents / retell_get_chat_agent / retell_create_chat_agent / retell_update_chat_agent / retell_delete_chat_agent / retell_get_chat_agent_versions / retell_publish_chat_agent
• Sessions: retell_list_chats / retell_get_chat / retell_create_chat / retell_update_chat / retell_create_chat_completion / retell_create_sms_chat / retell_end_chat

🔀 Conversation Flows (Node-based scripting)

• Flows: retell_list_conversation_flows / retell_get_conversation_flow / retell_create_conversation_flow / retell_update_conversation_flow / retell_delete_conversation_flow
• Components: list, get, create, update, delete for reusable conversation flow components.

⚙️ Utilities

• retell_create_batch_call / retell_create_batch_test
• retell_get_concurrency
• retell_get_mcp_tools

---

Architectural Notes

• Input Validation: Every tool validates its inputs dynamically using strict zod schemas. Invalid arguments return clean, detailed error messages.
• Type Safety: The entire codebase is thoroughly typed against the retell-sdk, eliminating any casting and masking TypeScript errors.
• Standard Protocol Compliance: All expected errors generate an MCP protocol isError: true payload, ensuring the LLM client correctly handles API rejections without crashing.