This MCP server provides an AI-assisted bridge between language models and WinDbg/CDB for Windows crash dump analysis and live debugging. It exposes tools that let an AI agent inspect crash dumps, connect to remote debugging sessions, enumerate and open dumps, and issue WinDbg commands in a structured MCP session. Typical workflows include analyzing a crash dump to identify where an exception originated, listing available dumps in a directory, and attaching to a running target for live debugging. The server is designed to work with MCP-enabled clients (such as Copilot, Claude Desktop, Windsurf, etc.) and can be accessed via standard I/O or the HTTP transport depending on your environment.

To use it, run the Python module as an MCP server (via the Python integration) or configure your MCP client to connect through the preferred transport. The included tools expose specific capabilities like listing dumps, opening a dump, closing a session, opening a remote debugging session, closing it, and running WinDbg commands. This enables you to perform analysis and debugging tasks in natural language while the backend translates commands to WinDbg/CDB operations.

Prerequisites:

• Windows OS with Debugging Tools for Windows (CDB/WinDbg) installed or available via the WinDbg tools package.
• Python 3.10 or higher installed on the system.
• An MCP-compatible client (e.g., Copilot, Claude Desktop, Windsurf) or the ability to connect via MCP transport.

Installation steps:

1. Install the MCP Windbg package from PyPI: pip install mcp-windbg

2. Verify Python environment and dependencies are available in your PATH. If needed, create and activate a virtual environment: python -m venv venv source venv/Scripts/activate # Windows

   or

   .\venv\Scripts\activate

3. Run the MCP Windbg server (example with default transport): python -m mcp_windbg

4. If you prefer HTTP transport for remote access, start with: python -m mcp_windbg --transport streamable-http --host 127.0.0.1 --port 8000

5. Connect a client (VS Code MCP integration or other MCP-enabled client) using the provided endpoint:

   • Standard I/O: use the default stdio channel as described in the server docs.
   • HTTP: http://127.0.0.1:8000/mcp

Notes and tips:

• The MCP Windbg server relies on having WinDbg/CDB available on the host. Ensure the symbol path is configured if you plan to analyze symbol-rich dumps.
• You can configure the Windows symbol path via the _NT_SYMBOL_PATH environment variable as shown in the VS Code example, or override it per-session as needed.
• If you encounter transport or connectivity issues, verify port bindings and firewall rules for the HTTP transport, and ensure the MCP client is aligned with the transport type you configured.
• The Tools section of the project provides a set of commands like list_windbg_dumps, open_windbg_dump, close_windbg_dump, open_windbg_remote, close_windbg_remote, and run_windbg_cmd. These are exposed through MCP endpoints and can be invoked via natural language prompts or direct API calls.
• In enterprise environments, confirm policy allowances for AI-assisted tooling and the use of debugging tools before enabling the MCP Windbg server.