Brings a deterministic pipeline state machine to Claude, exposing nine MCP tools for agent orchestration in VS Code with GitHub Copilot. You get pipeline_init, pipeline_reset, set_pipeline_mode for switching between supervised, assisted, and autopilot routing modes, plus satisfy_gate for approving stage transitions, skip_stage, and get_pipeline_status. The run_command tool executes shell commands directly from chat. Built around a fixed nine stage workflow (intent, PRD, architect, plan, implement, tester, review, closed) with 25 specialized agents. Useful if you're running complex AI driven development workflows and need predictable state transitions instead of LLM based routing. The Python state machine owns all handoffs, gates block progression until explicitly satisfied, and every transition is auditable.
claude mcp add --transport stdio io.github.saajunaid-junai-mcp -- uvx junai-mcp