Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.kataven.ai/llms.txt

Use this file to discover all available pages before exploring further.

You can run your entire Kataven account from an AI tool — create agents, configure tools, place calls, run campaigns — without writing API code yourself. The right integration path depends on whether your AI tool has a shell (can run python my_script.py) or is chat-only.

​
Pick your path

Your AI toolHas a shell?UseGuide
Claude CodeYesPython SDKManage with Claude Code →
Codex CLIYesPython SDKManage with Codex →
Cursor (composer / agent mode)YesPython SDKSame as Claude Code — Cursor runs python for you
Claude DesktopNoMCP serverManage with Claude Desktop →
ChatGPT DesktopNoMCP serverManage with ChatGPT →
OpenAI Agents SDK (programmatic)N/AMCP serverManage with ChatGPT § OpenAI Agents SDK
n8n / Zapier with an LLM stepNoMCP serverSame install as Claude Desktop
Custom agent (LangGraph, Agno, etc.)EitherEitherMCP is easier to plug in

​
Why two paths?

The Kataven Hub API is just HTTP. Anything that can speak HTTP can drive it.
  • AI tools with a shell install the Python SDK (pip install kataven) and let the AI write & run Python directly. No middleware. The SDK is the headless surface.
  • Chat-only AI tools install the Kataven MCP server (pip install kataven-mcp), which exposes every SDK verb as an MCP tool the AI can call from inside its chat loop. The MCP server is a thin wrapper around the SDK — same functionality, different protocol.
You don’t need both for the same client. Pick one.

​
Common setup (any path)

All four guides start the same way:
1

Mint an API key

hub.kataven.ai/settings → API Keys tab → Create API key. Copy the sk_live_… value once. Full walkthrough: Quickstart § 1.
2

Install the SDK or MCP server

Per-tool details in the guides below. Both ship as standard pip packages.
3

Configure your AI tool

Set KATAVEN_API_KEY=sk_live_… somewhere the tool can read it (env var for shell-based tools, MCP config file for chat-only tools).

​
What the AI can actually do

Every Hub API verb is reachable. Concretely:
  • kataven_list_agents, kataven_get_agent, kataven_create_agent, kataven_update_agent, kataven_delete_agent
  • kataven_attach_playbook / kataven_detach_playbook, kataven_attach_faq / kataven_detach_faq
  • kataven_list_tools, kataven_create_tool (custom tools)
  • kataven_list_phone_numbers, kataven_create_phone_number, kataven_update_phone_number
  • kataven_originate_call (place outbound test calls)
  • kataven_list_campaigns, kataven_create_campaign, kataven_start_campaign, kataven_pause_campaign, kataven_stop_campaign
  • kataven_list_widget_keys, kataven_create_widget_key
  • …and the full surface of the API reference.
A few admin endpoints stay out of reach by design (cost-cap edits, integration authoring, recording deletion) — those require a Hub UI session, and the server returns 403 to sk_live_ callers. See Authentication § Mint, list, revoke API keys.

​
Pick your guide

Claude Code

Anthropic’s terminal coding agent. Has a shell, uses the SDK directly, no MCP needed.

Claude Desktop

The chat-only Claude desktop app. Wires up via MCP server in claude_desktop_config.json.

ChatGPT

ChatGPT Desktop via MCP, or the OpenAI Agents SDK programmatically.

Codex CLI

OpenAI’s terminal coding agent. Has a shell, uses the SDK directly.