↓ Skip to main content
NTNINJA

dbgeng-mcp

Links#

Overview
#

dbgeng-mcp is a Model Context Protocol (MCP) server that bridges AI assistants to the Windows Debugging Engine (dbgeng.dll). Built with Python and a native C++ extension (pybind11), it exposes debugger functionality as MCP tools so that LLM-based agents can launch, attach to, and inspect processes through natural language. The project is currently alpha software under active development and is MIT licensed.

Features
#

Process Control
#

  • Launch a process under the debugger
  • Attach to a running process
  • Detach from the target
  • Restart the current debug session

Execution Control
#

  • Continue, break, and step execution
  • Run until a specific event occurs

State Inspection
#

  • Query debugger state
  • Retrieve debugger output and event log
  • Get stack traces

Memory & Commands
#

  • Read target memory
  • Execute arbitrary WinDbg commands

Extension Management
#

  • Load, unload, and list debugger extensions

Safety
#

  • Read-only mode
  • Automatic SDK discovery via Windows registry

Quick Start
#

1. Install from PyPI (recommended). dbgeng-mcp installs through uv/pip like any other MCP server (requires Python 3.10+ x64 and Debugging Tools for Windows):

# run on demand (fetched + cached on first launch)
uvx dbgeng-mcp

# or install the CLI persistently
uv tool install dbgeng-mcp

2. Example MCP client config (e.g. for Claude Code or Claude Desktop), using uvx:

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