liaohch3/claude-tap

claude-tap is a local proxy and trace viewer for AI coding agents.

The tool solves the problem of understanding what happens inside AI coding agents by intercepting and displaying the actual API traffic they generate. It works by running as a local proxy that captures requests and responses, then presents them in a web-based trace viewer where developers can inspect system prompts, conversation history, tool schemas, tool calls, streaming responses, token usage, and how context changes between requests. The viewer includes a structured diff modal to compare how the agent's context evolves across adjacent API calls.

The tool suits developers who need to debug or understand the behavior of AI coding agents, particularly those using Claude Code, various CLI tools from OpenAI, Google, Anthropic, and other providers, or building systems that integrate with these agents. It is especially valuable when you need to see the raw HTTP traffic and internal state that the agent is working with, rather than just the final output. The README does not compare it to alternative approaches.

The project shows active maintenance with regular updates to support new agent platforms and tools. Development includes ongoing expansion of agent compatibility across multiple vendors and frameworks. The codebase demonstrates responsiveness to user needs through documented guides and examples of how the tool integrates into broader workflows.