Agent Replay Debugger Mcp

Built for debugging agentic workflows step by step. Gives you three core operations: start_recording to capture a session, record_step to log each action with state snapshots, and replay_step to deterministically rerun individual steps or entire sequences. The big win is deterministic replay, so you can reproduce bugs without guessing what the agent was thinking three steps ago. Ships with EU AI Act compliance hooks baked in (audit trails, transparency logging), which matters if you're building anything that needs to explain itself later. Install via pip or uvx, runs over stdio. Reach for this when your agent chain breaks in production and you need more than print statements to figure out why.