Claude Code + TraceBug (MCP)

Claude Code bug reports:
capture it, let Claude Code fix it

Claude Code can fix a bug it can actually see. TraceBug captures the browser session — replay, console errors, the failed request, the exact repro steps — into one file Claude Code reads over a local MCP server, then reproduces and fixes it. Nothing is uploaded.

Connect TraceBug to Claude Code

Claude Code registers MCP servers with a single command — the fastest setup of any tool. Run it once and Claude can read every TraceBug export on your machine; there's also a built-in prompt, /tracebug:debug_bug_report.

Terminal — run once
claude mcp add tracebug --scope user -- npx -y tracebug mcp

Full walkthrough with troubleshooting: the Claude Code MCP setup guide.

What Claude Code reads from the report

The local server gives Claude Code read-tools over the captured evidence — console errors with stack traces, the failing network request and its response body, the millisecond repro timeline, and screenshots. It starts from the actual failing line instead of guessing from a description.

From root cause to a verified fix

Every report also embeds a generated failing test that replays the session and asserts the failure is gone. Claude Code runs it (red), patches the code, and re-runs until green — so the fix is verified, not just proposed.

Capture → Claude Code reads the evidence → fix → the test goes green.

Local, not cloud. The MCP server runs on your machine over stdio with zero network connections; the report is one offline file and secrets are masked at capture. Your bug data and your code never leave your machine. MIT open source.

See Claude Code fix a real bug
Unedited transcript: crash → root cause → verified fix.
Watch the proof