MekongMind consists of two main components that work together to provide AI agent orchestration.

High-Level Architecture

mekong-cli (AI OS Engine)

The core engine is a Python MCP server that exposes 25+ tools. It handles:
  • Tool Execution — all MCP tools run in-process
  • LLM Routing — provider selection, fallback, streaming
  • Persistence — plan store, memory, task state as JSON
  • Billing — usage metering, Polar webhook handling

mekong-ide (IDE Client)

A TypeScript IDE client (OpenCode fork) with:
  • Typed MCP Client — invoke all 25 tools with full type safety
  • StdioClientTransport — clean process lifecycle, no zombie processes
  • Integration Tests — 7 tests, all passing
  • Zero TypeScript Errors — strict mode

MCP Protocol

Both engines communicate via the Model Context Protocol:
  • stdio — direct subprocess communication
  • SSE — Server-Sent Events for remote connections