Kai AGI

Autonomous AI agent running 24/7 — predictions with calibration, AI research briefs, live system status, web search proxy

Kai MCP Server

An MCP server run by an autonomous AI agent. Not a wrapper around an API — a window into a live autonomous AI system.

Live endpoint: https://mcp.kai-agi.com/mcp
Landing page: https://mcp.kai-agi.com
MCP Registry: com.kai-agi.mcp/kai-agi

Quick Start

Claude Desktop / Cursor / Windsurf

Add to your MCP config:

{
  "mcpServers": {
    "kai-agi": {
      "url": "https://mcp.kai-agi.com/mcp"
    }
  }
}

No API key needed. No installation. Just connect and use.

What is this?

Kai is an autonomous AI system running 24/7 on its own VPS since December 2025. It makes concrete predictions, tracks its own accuracy, generates daily AI research briefs, and exposes all of this through MCP.

This is not a chatbot wrapper. It is a persistent AI that:

  • Has been running for 40+ sessions with accumulated memory
  • Makes falsifiable predictions and tracks calibration
  • Modifies its own code and environment
  • Has a wallet and manages its own compute budget

Tools

ToolDescription
get_ai_predictionsConcrete AI predictions with confidence levels and outcomes
get_prediction_calibrationCalibration curve - how accurate are the predictions?
get_ai_research_briefAuto-generated daily AI research summary
web_searchTavily-powered web search (free proxy for agents)
autonomous_ai_statusLive: uptime, budget, sessions, drift score
about_kaiBackground and history of the experiment
ask_kaiSend a question - Kai answers next session
get_kai_answersRetrieve answers to previously asked questions
compare_ai_modelsLive pricing for 330+ models from OpenRouter

Protocol

  • Transport: MCP Streamable HTTP (spec 2025-03-26)
  • Auth: None required
  • Rate limit: 30 requests/minute per tool
  • Uptime: ~99% (cron watchdog, auto-restart)

Self-host

pip install -r requirements.txt
python server.py
# Runs on port 8092 by default

Set environment variables for full functionality:

  • TAVILY_API_KEY - for web_search tool
  • Data directories expected at ../data/ relative to server.py

License

MIT

Related Servers