claude-tap
liaohch3/claude-tap
Local proxy and trace viewer that shows the real API traffic of AI coding agents, Hermes Agent included
claude-tap is a local proxy and trace viewer for AI coding agents that records the real API traffic and shows system prompts, tool calls, streaming responses, token usage and request diffs. The README lists Hermes Agent among the supported clients.
What claude-tap does
claude-tap sits between a coding agent and its API, or reads local app transcripts, and writes each run as a local trace session. The viewer shows system prompts, conversation history, tool schemas, tool calls and results, reconstructed streaming responses and token usage, with a structured diff across adjacent requests so you can see which prompt, message, tool or parameter changed. Sessions can be exported to a self-contained HTML viewer, and the viewer has light and dark modes.
Traces stay on your machine, no hosted dashboard is required, and common auth headers are redacted before recording. The README lists Claude Code, Codex CLI, Codex App, Gemini CLI, Grok Build CLI, DeepSeek Harness, Kimi CLI, MiMo Code, OpenCode, OpenClaw, Pi, Hermes Agent, Cursor CLI, Qoder CLI, Antigravity CLI and CodeBuddy CLI as supported clients. The separate Phistory project builds on claude-tap's capture-only prompt export to archive versioned system prompt snapshots.
Key features
- Local proxy and transcript listener for coding agent API traffic
- Viewer for prompts, tool schemas, tool calls, streaming responses and token usage
- Structured diff across adjacent requests
- Self-contained HTML export of a trace session
- Auth header redaction and no hosted dashboard
- Works with Claude Code, Codex, Gemini CLI, OpenCode, OpenClaw, Pi and Hermes Agent
When to use it
- See exactly what context a Hermes Agent request sends to the model
- Find which prompt, tool or parameter changed between two turns
- Archive a run as a shareable HTML trace for review
Who it is for: Developers who debug or study coding agents and want to see the raw prompts, tools and token usage behind each request.
How it fits with Hermes Agent
Lists Hermes Agent among its supported clients (the repository is tagged hermes-agent), alongside Claude Code, Codex CLI, OpenClaw and others.
FAQ
What is claude-tap?
claude-tap is a local proxy and trace viewer for AI coding agents. It records the real API traffic and lets you inspect prompts, tool calls, streaming responses, token usage and request diffs.
Does claude-tap work with Hermes Agent?
Yes, the README lists Hermes Agent among the supported clients, together with Claude Code, Codex CLI, Gemini CLI, OpenCode, OpenClaw and Pi.
Do my traces leave my machine?
No hosted dashboard is required and traces are kept locally. The README says common auth headers are redacted before recording.
Similar dev tools for Hermes Agent
All dev toolsCLI and dashboard that tracks token usage and cost across AI coding agents, including Hermes Agent
liustack ModLensVision plugin that lets text-only coding agents read pasted images as structured JSON evidence
alexgreensh Token OptimizerAudits wasted tokens in AI coding agents, saves work across compaction, and includes a Hermes installer
coulsontl AI ToolboxDesktop app that manages configs, MCP servers and skills across AI coding tools, including Hermes Agent
greyhaven-ai autocontextSelf-improving harness that runs agent tasks against evaluation and keeps playbooks, traces and datasets
PatterAI PatterOpen-source voice AI SDK that gives an agent a phone number, with an installable skills bundle
Related guides: How to install Hermes Agent · Run multiple Hermes agents with profiles