elevenlabs-mcp-experiments

A small Model Context Protocol server for poking at ElevenLabs workspace data from inside Claude Desktop or any other MCP client.

I'm Auskin Immanuel. ElevenLabs Agents talks to external systems through tool calls and through MCP. I write tool definitions every week at work; this is the same idea from the server side.

What it exposes

Two tools, both read-only against the ElevenLabs workspace:

• list_agents: returns every agent in the workspace with its ID and name

• get_eval_runs: given an agent ID, pulls eval criteria results off its recent conversations and returns a compact summary per conversation

Read-only on purpose. Mutating things by accident from a chat client is not a fun way to learn the protocol.

Related MCP server: Pankow MCP

Run

python -m venv venv
source venv/bin/activate
pip install -e .
cp .env.example .env   # add ELEVENLABS_API_KEY
python server.py

To wire into Claude Desktop, add this to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "elevenlabs": {
      "command": "/absolute/path/to/venv/bin/python",
      "args": ["/absolute/path/to/server.py"]
    }
  }
}

Restart Claude Desktop. Ask Claude "list my ElevenLabs agents." The tool call should come back with real data.

What I'm trying to learn

• How MCP servers actually expose tools vs the webhook-tool pattern I use at work

• Where the protocol surfaces are (resources? prompts? tools only?)

• What it would take to ship a workspace-mutating tool (create agent, update prompt) safely

What this isn't

• Not a maintained package. There's no test suite, no CI, no release.

• Not auth-protected. Read your .env and don't run this on a shared machine.

Reach me

LinkedIn