The Kataven MCP server exposes every Hub API verb as a Model Context Protocol tool. Any MCP-aware client (Claude Desktop, Cursor’s chat, ChatGPT Desktop, n8n, OpenAI Agents SDK, etc.) can then manage your agents, tools, telephony, and more — without the user writing code.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 probably don’t need MCP if you’re using Claude Code.
Claude Code can write Python and run it directly against the
Python SDK, which is the headless surface.
MCP is for chat-only agent contexts that don’t have a shell.
Install
Configure
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json (Mac):
Cursor
~/.cursor/mcp.json:
Generic MCP (stdio)
What gets exposed
Every SDK verb becomes an MCP tool, prefixedkataven_:
kataven_list_agents,kataven_get_agent,kataven_create_agent,kataven_update_agent,kataven_delete_agentkataven_attach_playbook,kataven_detach_playbook,kataven_attach_faq,kataven_detach_faqkataven_list_tools,kataven_create_tool, …kataven_list_phone_numbers,kataven_create_phone_number, …kataven_originate_callkataven_create_campaign,kataven_start_campaign, …- … and so on for every resource.
Example sessions
“Show me all my agents and which phone numbers point to them.” Claude callskataven_list_agents, thenkataven_list_phone_numbersfor each, and presents the join.
“Create a ‘Tier 2 Support’ agent with a friendlier tone than the existing ‘Support’ agent.” Claude reads the existing agent viakataven_get_agent, derives a revised prompt, callskataven_create_agentwith the new config.
“Pause every running campaign so I can deploy.” Claude callskataven_list_campaigns, filters tostatus=running, and callskataven_pause_campaignon each.