Debugging AI agents can be challenging due to issues like non-reproducible tool failures, dependencies on external APIs, and the need for frequent reruns. Time Travel addresses these problems by acting as a transparent proxy for MCP agents. It records complete MCP sessions, capturing every tool call with detailed input, output, and timing metadata. This allows developers to deterministically replay sessions, step through them interactively for deep debugging, modify inputs/outputs on the fly, or quickly review a session summary, all without needing the real MCP server or an internet connection.
Key Features
011 GitHub stars
02Deterministic replay of recorded tool sessions
03Ability to modify inputs and override outputs during debugging
04Transparent MCP proxy for recording sessions
05Interactive step debugger to inspect and modify calls
06Works offline, no changes required to existing MCP servers
