Show HN: Callwitness – record what your AI agent's tools return (github.com)

Callwitness has launched a new tool designed to record the activity of AI agents by capturing every byte of data exchanged between the agent and its tools. This proxy recorder not only relays all data unchanged but also implements hash-chaining to maintain an unalterable log of transactions. The tool is versatile, requiring no setup or configuration, and is compatible with any MCP server running Python 3.8 or higher. Its capability to document responses across multiple public servers highlights significant potential for analyzing agent behavior and improving security protocols. The importance of Callwitness lies in its ability to generate real-time stats and logs that can inform safety and enforcement measures in AI deployments. As security guidelines evolve, especially regarding unmeasured agent failures, having precise data on agent outputs becomes crucial. Callwitness not only records but also assesses the data for anomalies and outputs insights on usage, offering developers a clear view of their agents' operational metrics. This focus on data-driven decision-making fosters accountability and transparency in AI applications, addressing a critical gap in current security practices.