Lapdog wraps your claude code or codex session and captures tool calls, LLM calls, and model reasoning for you to inspect.

Loved, it help me figure out how to improve my MCP’s tool descriptions so my agent wouldn’t take so many tries to get the right tool args