A Model Context Protocol server providing tools to analyze Windows crash dumps using WinDBG/CDB.

This MCP server integrates with CDB to enable AI models to analyze Windows crash dumps.

• Primarily, a tool that enables AI to interact with WinDBG.
• The whole "magic" is giving LLMs the ability to execute debugger commands. Used creatively, this is quite powerful and a big productivity improvement.

This means, that this is:

• A bridge connecting LLMs (AI) with WinDBG (CDB) for assisted crash dump analysis.
• A way to get immediate first-level triage analysis, useful for categorizing crash dumps or auto-analyzing simple cases.
• A platform for natural language-based "vibe" analysis, allowing you to ask the LLM to inspect specific areas:
  • Examples:
    • "Show me the call stack with k and explain what might be causing this access violation"
    • "Execute !peb and tell me if there are any environment variables that might affect this crash"
    • "Examine frame 3 and analyze the parameters passed to this function"
    • "Use dx -r2 on this object and explain its state" (equivalent to dx -r2 ((MyClass*)0x12345678))
    • "Analyze this heap address with !heap -p -a 0xABCD1234 and check for buffer overflow"
    • "Run .ecxr followed by k and explain the exception's root cause"
    • "Check for timing issues in the thread pool with !runaway and !threads"
    • "Examine memory around this address with db/dw/dd to identify corruption patterns"
    • ...and many other analytical approaches based on your specific crash scenario

• A magical solution that automatically fixes all issues.
• A full-featured product with custom AI. Instead, it's a simple Python wrapper around CDB that relies on the LLM's WinDBG expertise, best complemented by your own domain knowledge.

I've written about the whole journey in blog.

• The Future of Crash Analysis: AI Meets WinDBG

• Python 3.10 or higher
• Windows operating system with Debugging Tools for Windows installed.
  • This is part of the Windows SDK.
• A LLM supporting Model Context Protocol.
  • I have tested with Claude 3.7 Sonnet through GitHub Copilot and I am pretty statisfied with the results.
  • For GitHub Copilot, requires Model Context Protocol in Chat feature enabled.
  • See Extending Copilot Chat with the Model Context Protocol (MCP).

1. Clone the repository:

git clone https://github.com/svnscha/mcp-windbg.git
cd mcp-windbg

2. Create and activate a virtual environment:

python -m venv .venv
.\.venv\Scripts\activate

3. Install the package in development mode:

pip install -e .

4. Install test dependencies:

pip install -e ".[test]"

To integrate this MCP server with Visual Studio Code:

1. Create a .vscode/mcp.json file in your workspace with the following configuration:

{
    "servers": {
        "mcp_server_windbg": {
            "type": "stdio",
            "command": "${workspaceFolder}/.venv/Scripts/python",
            "args": [
                "-m",
                "mcp_server_windbg"
            ],
            "env": {
                "_NT_SYMBOL_PATH": "SRV*C:\\Symbols*https://msdl.microsoft.com/download/symbols"
            }
        },
    }
}

Alternatively, edit your user settings to enable it globally (independent of workspace). Once added and with Model Context Protocol in Chat feature enabled, the tools from this model context protocol server are available in Agent mode.

That's how it should look like:

If integrated through Copilot, you don't need this. The IDE will auto-start the MCP.

Start the server using the module command:

python -m mcp_server_windbg

python -m mcp_server_windbg [options]

Available options:

• --cdb-path CDB_PATH: Custom path to cdb.exe
• --symbols-path SYMBOLS_PATH: Custom symbols path
• --timeout TIMEOUT: Command timeout in seconds (default: 30)
• --verbose: Enable verbose output

2. Customize the configuration as needed:
   • Adjust the Python interpreter path if needed
   • Set custom paths for CDB by adding "--cdb-path": "C:\\path\\to\\cdb.exe" to the args array
   • Set the symbol path environment variable as shown above, or add "--symbols-path" to the args

Once the server is configured in VS Code:

1. Enable MCP in Chat feature in Copilot settings
2. The MCP server will appear in Copilot's available tools
3. The WinDBG analysis capabilities will be accessible through Copilot's interface
4. You can now analyze crash dumps directly through Copilot using natural language queries

This server provides the following tools:

• open_windbg_dump: Analyze a Windows crash dump file using common WinDBG commands
• run_windbg_cmd: Execute a specific WinDBG command on the loaded crash dump
• list_windbg_dumps: List Windows crash dump (.dmp) files in the specified directory.
• close_windbg_dump: Unload a crash dump and release resources

To run the tests:

pytest

If you get a "CDB executable not found" error, make sure:

1. WinDBG/CDB is installed on your system
2. The CDB executable is in your system PATH, or
3. You specify the path using the --cdb-path option

For proper crash analysis, set up your symbol path:

1. Use the --symbols-path parameter, or
2. Set the _NT_SYMBOL_PATH environment variable

SRV*C:\Symbols*https://msdl.microsoft.com/download/symbols

MIT