Krometrail

Appearance

Using Krometrail with Claude Code ​

Krometrail gives Claude Code runtime debugging — it can set breakpoints, inspect variables, and trace execution to find bugs.

Add via the CLI or edit the project .mcp.json directly:

bash
claude mcp add --transport stdio --scope project krometrail -- krometrail --mcp

Or add to .mcp.json in your project root (shared with your team):

json
{
  "mcpServers": {
    "krometrail": {
      "command": "krometrail",
      "args": ["--mcp"]
    }
  }
}
json
{
  "mcpServers": {
    "krometrail": {
      "command": "npx",
      "args": ["krometrail@latest", "--mcp"]
    }
  }
}
json
{
  "mcpServers": {
    "krometrail": {
      "command": "bunx",
      "args": ["krometrail@latest", "--mcp"]
    }
  }
}

Claude discovers the debug_* tools automatically — no CLAUDE.md changes needed.

Focused tool sets ​

If you only need debugging or only need browser observation, add a filtered server:

bash
# Debug tools only
claude mcp add --transport stdio --scope project krometrail-debug -- krometrail --mcp --tools=debug

# Browser tools only
claude mcp add --transport stdio --scope project krometrail-browser -- krometrail --mcp --tools=browser

Setup: CLI with Skill ​

The CLI provides the same debugging capabilities as the MCP tools, accessed through bash commands. Some setups prefer CLI tools for their transparency and scriptability.

Install the krometrail skill:

bash
npx skills add nklisch/krometrail --skill krometrail-debug krometrail-chrome

Verification ​

  1. Start Claude Code
  2. Ask: "What debug tools do you have available?"
  3. Claude should list the debug_* tools (MCP) or describe the CLI commands (skill)

Example Workflow ​

Ask Claude Code:

The test_gold_discount test is failing with an assertion error. Use the debugger to find the root cause.

Claude Code will:

  1. Identify the test file and set a breakpoint at the assertion
  2. Launch the test under the debugger: debug_launch with pytest tests/test_discount.py::test_gold_discount
  3. Continue to the breakpoint: debug_continue
  4. Inspect variables: debug_evaluate on tier_multipliers, discount, etc.
  5. Trace back to find the incorrect value
  6. Report the root cause with the specific line and variable that is wrong

Tips ​

  • MCP path is zero-config — Claude discovers tools automatically from the server
  • CLI path exposes the same tools as bash commands — transparent, scriptable, and preferred by some agent setups
  • Let Claude choose breakpoints — it knows the code better after reading it
  • Conditional breakpoints are powerful for loops: krometrail debug break app.py:25 when discount < 0
  • The viewport is compact (~400 tokens per stop) so Claude can take many debug steps without exhausting context
  • Framework auto-detection works for pytest, jest, go test, Django, Flask, etc.

Checking Installation ​

bash
krometrail doctor

This shows which language adapters are installed and their versions.