This server wraps Claude Code, Codex, Gemini CLI, Forge, and OpenCode so you can orchestrate multiple AI agents from within Claude Desktop or other MCP clients. It exposes run, wait, kill, and result tools that launch background processes with permissions bypassed, return immediately with a PID, and let you poll or block until completion. The real win is parallel execution: kick off three refactoring tasks with different models, move on to something else, then collect results when they finish. It also supports session IDs for context reuse across runs, which saves prompt caching costs when you're iterating on large codebases. You'll need each CLI installed and authenticated locally before the server can invoke them non-interactively.
claude mcp add --transport stdio mkxultra-ai-cli-mcp uvx ai-cli-mcp