MCP Server for WinDbg Crash Analysis
A Model Context Protocol server that bridges AI models with WinDbg for crash dump analysis and remote debugging.
<!-- mcp-name: io.github.svnscha/mcp-windbg -->Overview
This MCP server integrates with CDB to enable AI models to analyze Windows crash dumps and connect to remote debugging sessions using WinDbg/CDB.
What is this?
An AI-powered tool that bridges LLMs with WinDbg for crash dump analysis and live debugging. Execute debugger commands through natural language queries like "Show me the call stack and explain this access violation".
What This is Not
Not a magical auto-fix solution. It's a Python wrapper around CDB that leverages LLM knowledge to assist with debugging.
Usage Modes
- Crash Dump Analysis: Examine Windows crash dumps
- Live Debugging: Connect to remote debugging targets
- Directory Analysis: Process multiple dumps for patterns
Quick Start
Prerequisites
- Windows with Debugging Tools for Windows or WinDbg from Microsoft Store.
- Python 3.10 or higher
- Any MCP-compatible client (GitHub Copilot, Claude Desktop, Cline, Cursor, Windsurf etc.)
- Configure MCP server in your chosen client
TIP
In enterprise environments, MCP server usage might be restricted by organizational policies. Check with your IT team about AI tool usage and ensure you have the necessary permissions before proceeding.
Installation
pip install mcp-windbg
Transport Options
The MCP server supports multiple transport protocols:
| Transport | Description | Use Case |
|---|---|---|
stdio (default) | Standard input/output | Local MCP clients like VS Code, Claude Desktop |
streamable-http | Streamable HTTP | Modern HTTP clients with bidirectional streaming |
Starting with Different Transports
Standard I/O (default):
mcp-windbg
# or explicitly
mcp-windbg --transport stdio
Streamable HTTP:
mcp-windbg --transport streamable-http --host 127.0.0.1 --port 8000
Endpoint: http://127.0.0.1:8000/mcp
Command Line Options
--transport {stdio,streamable-http} Transport protocol (default: stdio)
--host HOST HTTP server host (default: 127.0.0.1)
--port PORT HTTP server port (default: 8000)
--cdb-path PATH Custom path to cdb.exe
--symbols-path PATH Custom symbols path
--filter-script PATH Python script with process_input/process_output tool text hooks
--timeout SECONDS Command timeout (default: 30)
--verbose Enable verbose output
Filter Script Hooks
Use --filter-script to load a small Python helper that rewrites tool text only (for example, to redact PII) without seeing the full MCP JSON-RPC envelope:
mcp-windbg --filter-script C:\filters\pii_redaction.py
The script defines process_input and/or process_output callbacks and runs in-process, so treat it as trusted code. See Redact sensitive data for the callback contract and a worked example.
Configuration
mcp-windbg works with any MCP client. Two common setups are below; see the client configuration guide for Claude Desktop, Copilot CLI, HTTP, and from-source.
VS Code (GitHub Copilot) - press F1 and select MCP: Open User Configuration to enable it in every workspace:
{
"servers": {
"mcp_windbg": {
"type": "stdio",
"command": "python",
"args": ["-m", "mcp_windbg"],
"env": {
"_NT_SYMBOL_PATH": "SRV*C:\\Symbols*https://msdl.microsoft.com/download/symbols"
}
}
}
}
Claude Code - register the server from the command line:
claude mcp add mcp-windbg -s user -e _NT_SYMBOL_PATH="SRV*C:\Symbols*https://msdl.microsoft.com/download/symbols" -- python -m mcp_windbg
Prefer not to install the package? Replace python -m mcp_windbg with uvx --from git+https://github.com/svnscha/mcp-windbg mcp-windbg in either setup to fetch and run the server on demand.
Once configured, restart your MCP client and start debugging:
Analyze the crash dump at C:\dumps\app.dmp
MCP Compatibility
This server implements the Model Context Protocol (MCP), making it compatible with any MCP-enabled client:
The beauty of MCP is that you write the server once, and it works everywhere. Choose your favorite AI assistant!
Tools
| Tool | Purpose | Use Case |
|---|---|---|
list_windbg_dumps | List crash dump files | Discovery and batch analysis |
open_windbg_dump | Analyze crash dumps | Initial crash dump analysis |
close_windbg_dump | Cleanup dump sessions | Resource management |
open_windbg_remote | Connect to remote debugging | Live debugging sessions |
close_windbg_remote | Cleanup remote sessions | Resource management |
run_windbg_cmd | Execute WinDbg commands | Custom analysis and investigation |
send_ctrl_break | Break into a running target | Interrupt execution during live debugging |
Documentation
| Topic | Description |
|---|---|
| Getting Started | Quick setup and first crash dump analysis |
| Use cases | Analyze a dump, debug a remote target, triage many dumps |
| Command-line options | Every CLI flag, transports, and filter hooks |
| Tools Reference | The MCP tools and their parameters |
| Client configuration | VS Code, Claude Desktop, Copilot CLI, pip, and source |
| Troubleshooting | Common issues and solutions |
Examples
Crash Dump Analysis
Analyze this heap address with !heap -p -a 0xABCD1234 and check for buffer overflow"
Execute !peb and tell me if there are any environment variables that might affect this crash"
Run .ecxr followed by k and explain the exception's root cause"
Remote Debugging
"Connect to tcp:Port=5005,Server=192.168.0.100 and show me the current thread state"
"Send CTRL+BREAK to the live session, then dump all thread stacks with ~*k"
"Check for timing issues in the thread pool with !runaway and !threads"
"Show me all threads with ~*k and identify which one is causing the hang"
Blog
Read about the development journey: The Future of Crash Analysis: AI Meets WinDbg
Links
Star History
License
MIT