debugger-mcp

MCP server that exposes debugger tools to AI clients (Cursor, Claude Code, etc.): launch, breakpoints, continue, step, evaluate, stack, variables, probe, stop. Supports Node.js (CDP) and Python (DAP). The debugger runtime (top-coder-ai-skills-debugger) is included as a dependency — no separate installation.

Install

One install is enough (the debugger library is pulled in automatically):

pip install debugger-mcp
# or
uv add debugger-mcp
# or
poetry add debugger-mcp

Requires: Python ≥3.11. Nothing else to install.

Global

Use without adding the package to a project:

• pipx (install once, run from anywhere):

  pipx install debugger-mcp
  debugger-mcp
  

  Then in MCP config use "command": "debugger-mcp" (pipx puts it on PATH).

• uvx (run on demand, no install — like npx):

  uvx debugger-mcp
  

  In MCP config: "command": "uvx", "args": ["debugger-mcp"].

Run

Stdio transport (e.g. for Cursor MCP). Configure in your MCP client (e.g. Cursor) using one of these:

Option A — pipx (after pipx install debugger-mcp; use if debugger-mcp is on your PATH):

{
  "mcpServers": {
    "debugger": {
      "command": "debugger-mcp"
    }
  }
}

Option B — uvx (no install; runs on demand, like npx):

{
  "mcpServers": {
    "debugger": {
      "command": "uvx",
      "args": ["debugger-mcp"]
    }
  }
}

Option C — in-project (if you added the package to the project with uv add / poetry add / pip install):

{
  "mcpServers": {
    "debugger": {
      "command": "python3",
      "args": ["-m", "debugger_mcp"]
    }
  }
}

If using Option A and the app doesn’t see debugger-mcp on PATH, use the full path (e.g. ~/.local/bin/debugger-mcp on Linux/macOS) or use Option B (uvx) instead.

Tools

See the repo and ARCHITECTURE.md for protocol and usage details.