LLM Interceptor
A cross-platform man-in-the-middle proxy that intercepts, logs and analyzes the traffic between AI coding tools such as Claude Code, Cursor, Codex or OpenCode and their backend language-model APIs.
LLM Interceptor sits between an AI coding agent and its provider's API to capture every prompt sent and response received, including streamed server-sent-event responses, across providers such as Anthropic, OpenAI's Chat Completions and Responses APIs, Google, Groq, Together and Mistral. It offers an interactive watch mode for continuous capture with session management, and a one-shot run mode that wraps a command such as a Claude Code or Codex invocation for fully automated, non-interactive capture suited to scripted experiments; captured API keys and other sensitive data are automatically masked in the logs, and session data is merged and split automatically.
It bundles an agent skill, installable through the open skills CLI, that teaches a compatible AI agent how to capture traffic and analyze existing trace directories, including token usage, latency and tool calls, and can even bootstrap the interceptor itself if it is not yet installed.
It is written in Python and installable with uv or pip.