Skip to main content
You don’t need to write code. The Kataven MCP server is a small program you install once on your laptop. After that, your favorite AI app — Claude Desktop, ChatGPT, Cursor, Windsurf, Zed, VS Code’s Copilot, JetBrains, Warp, Goose, n8n, Raycast — can manage your Kataven voice agents by chatting. Ask Claude “create a sales agent and place a test call to my cell” and it does. No SDK, no curl, no Python.

Who this page is for

You are…Use this page if…
A non-developer founder, ops lead, or marketer running KatavenYou want to ask Claude / ChatGPT to do things (“pause the running campaign”, “show me numbers without an agent”) instead of clicking around the dashboard
An ops engineer automating Kataven from a chat surfaceYou want one config blob you can paste into every team member’s AI app
A solo developer using Claude Desktop, Cursor, or ZedYou’d rather chat than install Python and write SDK calls
If you’re using Claude Code or Codex CLI (terminal coding agents that have a shell), you don’t need MCP — those tools install the Python SDK directly. See Manage with Claude Code or Manage with Codex. MCP is for everything else.

Two ways MCP clients connect

kataven-mcp runs locally on your machine over stdio (a local pipe). Most AI apps connect to it directly. A few — ChatGPT, Claude.ai web, n8n — only accept remote MCP URLs over HTTPS, so you need a one-line tunnel to expose your local server publicly.
AI appConnectionDifficultyGuide
Claude DesktopDirect stdio🟢 paste JSON, restartSetup →
CursorDirect stdio🟢 paste JSON, restartSetup →
WindsurfDirect stdio🟢 UI button + JSONSetup →
ZedDirect stdio🟢 settings.json editSetup →
VS Code (Copilot agent)Direct stdio🟢 Command PaletteSetup →
JetBrains AI AssistantDirect stdio🟢 UI formSetup →
Warp TerminalDirect stdio🟢 UI buttonSetup →
GooseDirect stdio🟢 goose configureSetup →
Raycast AIDirect stdio🟢 install MCP extensionSetup →
ChatGPT (Desktop + web + mobile)Remote HTTPS only🟡 needs tunnelSetup →
Claude.ai web (Pro+)Remote HTTPS only🟡 needs tunnelSetup →
n8nRemote SSE only🟡 needs tunnelSetup →
Building your own AI agent (LangGraph, Agno, AutoGen, OpenAI Agents SDK)? Stdio works directly — see Programmatic.

The universal pattern

Every client follows the same three steps:
1

Install the MCP server

pip install kataven-mcp
Don’t have Python? On macOS run brew install python first; on Windows install official Python 3.10+. Full walkthrough: Install →.
2

Mint an API key

hub.kataven.ai/settingsAPI Keys tab → Create API key. Copy the sk_live_… value once.
3

Configure your AI app

Pick your client from the table above. Each guide is one page; the JSON entry is identical everywhere — only the file path or UI button differs.
After step 3, restart the AI app and ask it to do something Kataven-related. The Kataven tools (kataven_list_agents, kataven_create_agent, kataven_originate_call, …) appear in its tool picker.

What the AI can do

Real prompts that work once kataven-mcp is wired up:
“Show me every agent and which phone number it’s pinned to. Anything pointing nowhere?”
“Create a ‘Tier 2 Support’ agent with a friendlier tone than the existing ‘Support’ agent. Attach the general-faqs playbook.”
“Place a test call from +12025550123 to my cell +14155550100 using the Front Desk agent. Show me the live transcript link.”
“Pause every running campaign — we’re deploying a fix.”
Full list of exposed tools: see Install § What gets exposed.

Important reading before you go further

  • Tunneling — required if you’re using ChatGPT, Claude.ai web, or n8n.
  • Security — your sk_live_ is full account scope; treat it like a CI deploy key.
  • Troubleshooting — what to do when tools don’t appear.
  • Programmatic — OpenAI Agents SDK, LangGraph, Agno, AutoGen.

Other ways to manage Kataven from AI