You can manage Kataven entirely through Claude. Two flavors depending on where you’re using Claude.Documentation Index
Fetch the complete documentation index at: https://katavenai.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
Option A — Claude Code (or any agent with a shell)
Claude Code can write Python and execute it. Install the SDK once, configure the env vars, and ask Claude to do the rest.“List all my agents and tell me which ones are active. Then create a new agent called ‘Sales SDR’ that knows about our pricing tiers.”Claude will read the SDK docs, write the Python, and run it. No MCP required — the SDK is the headless surface.
Option B — Claude Desktop / Cursor / any chat-only agent
Chat-only Claude has no shell. To drive Kataven from there, install the MCP server, which exposes every SDK verb as a Claude-callable tool.Install
Configure Claude Desktop
Edit~/Library/Application Support/Claude/claude_desktop_config.json:
kataven_list_agents,
kataven_create_agent, kataven_originate_call, etc.) appear in the
tool list.
Then ask away
“Show me all phone numbers I’ve registered, and tell me which ones aren’t pinned to an agent.”
“Create an agent called ‘Front Desk’ with a friendly greeting, then attach the ‘general-faqs’ playbook to it.”
“Place a test call from +12025550123 to my cell +14155550100 using the Front Desk agent.”
Which to choose?
| You’re working in… | Use |
|---|---|
| Claude Code | SDK (no MCP needed — Claude has a shell) |
| Cursor with composer / agent mode | SDK |
| Claude Desktop, ChatGPT Desktop, mobile chat | MCP |
| n8n / Zapier-style workflow with an LLM step | MCP |
| Custom agent (LangGraph, Agno, etc.) | Either; MCP is easier to plug in |