Claude Inspector is an Electron desktop application by Kangraemin, designed as a local Man-in-the-Middle proxy for the Claude Code CLI. It functions as a Claude Code Prompt Mechanism Visualizer, providing developers with a dedicated tool to inspect and debug AI agent interactions. The application intercepts network traffic between the Claude Code CLI and the Anthropic API, offering detailed insights into API payloads, session flows, and token usage. It aims to enhance understanding of how AI agents operate and consume resources.

What does Claude Inspector connect to?

Claude Inspector connects directly to the Claude Code CLI, acting as a local Man-in-the-Middle proxy. It intercepts the network traffic that flows between the Claude Code CLI and the api.anthropic.com endpoint. This allows it to observe and analyze the HTTP/HTTPS requests and responses exchanged during AI agent interactions. Its primary role is to provide a visualization and debugging layer for the Claude Code CLI's communication with the Anthropic API.

Core features and tools

Claude Inspector provides a graphical user interface with several tabs for inspection. The AI Analysis tab allows users to initiate an "Analyze Session Flow". Other tabs include Request, Response, and Anatomy, offering detailed views of intercepted traffic. Users can click any "Request #N badge" to navigate directly to that request's tool_use within the JSON tree, facilitating precise examination of AI agent interactions.

Who is it for?

Claude Inspector is for developers who use the Claude Code CLI and require a detailed understanding and debugging capability for their AI agent interactions. It helps in visualizing prompt mechanisms, analyzing API payloads, and optimizing token consumption. The tool assists those who need to see exactly how CLAUDE.md and MCP tools influence Claude Code's operations. Skip this if you are not a developer using the Claude Code CLI for AI agent development.

How to get started

No credentials are required directly by Claude Inspector itself; it observes traffic from the Claude Code CLI, which handles its own authentication. Setup involves configuring the Claude Code CLI to direct its API requests through the inspector's local proxy address, for example, by setting ANTHROPIC_BASE_URL to "http://localhost:9090".

Use cases

• Real-time inspection of API payloads exchanged between Claude Code CLI and the Anthropic API.
• AI-powered session flow analysis, including summaries, MCP dynamic loading chains, and sub-agent patterns.
• Debugging prompt mechanisms, CLAUDE.md usage, and token consumption during Claude Code operations.

Claude Inspector by Kangraemin serves as a specialized local MITM proxy for developers working with the Claude Code CLI. It offers a graphical interface to visualize, inspect, and debug AI agent interactions, providing insights into API requests, responses, and token usage. The tool is valuable for understanding prompt mechanisms and optimizing AI agent behavior.