TwelveLabs MCP Server

This project is not affiliated with TwelveLabs (the video understanding platform). It is an MCP server for the ElevenLabs Conversational AI API — voice agents, knowledge bases, conversations, and TTS. The name "TwelveLabs" is the project codename only.

An enhanced MCP (Model Context Protocol) server that gives Claude direct access to the ElevenLabs Conversational AI API -- including capabilities the official ElevenLabs MCP connector doesn't expose.

What this covers

• Agents -- list, get full config (including LLM model + temperature), update prompt/settings, manage webhooks
• Knowledge Base -- list, read full content, add text/URL docs, delete docs
• Conversations -- list, get full transcript + data_collection analysis
• Voices -- list and search available voices

Prerequisites

• Node.js >= 18
• ElevenLabs API key -- get one at elevenlabs.io

Quick Start

Option A: npx (no clone needed)

ELEVENLABS_API_KEY=your-key npx twelvelabs-mcp-server

Option B: Clone and build

git clone https://github.com/micro-JAY/twelvelabs-mcp-server.git
cd twelvelabs-mcp-server
npm install --include=dev
npm run build

Add to Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "twelvelabs": {
      "command": "node",
      "args": ["/path/to/twelvelabs-mcp-server/dist/index.js"],
      "env": {
        "ELEVENLABS_API_KEY": "your-api-key-here"
      }
    }
  }
}

Then restart Claude Desktop. You should see "twelvelabs" in the available tools panel.

Use it

In any Claude conversation:

"List my agents" "Get the full config for agent_1001kjbh14xce5kbreh55ydf4rae" "Show me the last 10 conversations for that agent" "Get the transcript and analysis for conversation_xyz"

Tool Reference

Agent Tools

| Tool | What it does | |------|-------------| | twelvelabs_list_agents | List all agents in your account | | twelvelabs_get_agent | Full config: LLM model, voice, KB docs, webhook, prompt preview | | twelvelabs_get_agent_prompt | Full untruncated system prompt text | | twelvelabs_update_agent_prompt | Replace system prompt and/or first message | | twelvelabs_update_agent_settings | Change temperature, voice, TTS settings, language | | twelvelabs_list_agent_conversations | Recent conversations for one agent | | twelvelabs_get_agent_webhook | Current webhook URL | | twelvelabs_set_agent_webhook | Set or remove webhook URL |

Knowledge Base Tools

| Tool | What it does | |------|-------------| | twelvelabs_list_kb_docs | All KB docs with size and creation date | | twelvelabs_get_kb_doc | Full extracted text content of a doc | | twelvelabs_add_kb_text | Add a plain text document | | twelvelabs_add_kb_url | Add a web page by URL | | twelvelabs_delete_kb_doc | Permanently delete a doc |

Conversation Tools

| Tool | What it does | |------|-------------| | twelvelabs_list_conversations | Recent conversations (optionally filtered by agent) | | twelvelabs_get_conversation | Full transcript + data_collection + evaluation_criteria |

Voice Tools

| Tool | What it does | |------|-------------| | twelvelabs_list_voices | Browse available voices with IDs and labels |

Security

• Input validation -- all tool parameters are validated with Zod schemas before reaching the API
• HTTPS enforcement -- webhook URLs are validated to use HTTPS only
• Path traversal protection -- document and agent IDs are sanitized to prevent path traversal
• Error sanitization -- internal errors are caught and returned as safe MCP error responses; raw stack traces are never exposed to the client

Development

npm run dev   # watch mode with tsx -- auto-reloads on file changes
npm run build # compile to dist/

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feat/my-feature)
3. Commit your changes (git commit -m 'feat: add my feature')
4. Push to the branch (git push origin feat/my-feature)
5. Open a Pull Request

License

MIT -- see LICENSE for details.

Architecture Note

All tools share a single Axios client instance (src/client.ts) initialised from ELEVENLABS_API_KEY at startup. Tool implementations live in src/tools/ -- one file per domain -- and are registered onto the McpServer in src/index.ts. The server communicates over stdio transport (stdin/stdout), which is why all debug logging uses console.error (stderr), not console.log.