Analyze a crash dump¶
You have a .dmp file and want to know what went wrong. Point the model at it:
This calls open_windbg_dump, which opens the dump
in cdb.exe and runs the standard triage commands in one pass:
- the crash information (
.lastevent) - the automated analysis (
!analyze -v) - the stack trace, loaded modules, and threads
The session stays open, so every follow-up question reuses it instead of reloading the dump.
Ask follow-up questions¶
Once the dump is open, keep going in plain language. Each request becomes a
run_windbg_cmd call against the same session:
Show the call stack with kb and explain the access violation
Run .ecxr then u to disassemble around the faulting instruction
Dump the locals for frame 2
Check the heap around 0x1f2a0040 with !heap -p -a 0x1f2a0040
List the loaded modules with lm and flag anything unsigned
You do not have to memorize commands, describe what you want and let the model choose. The Tools reference lists the commands that come up most often if you want to be specific.
Symbols make or break this¶
Readable stack traces need symbols. The Microsoft symbol server is configured through
_NT_SYMBOL_PATH in your client config (see Getting started). For
your own binaries, put the matching .pdb files next to the dump, the server automatically
adds the dump's own directory to the symbol path. To add more locations for a single call:
That maps to the per-call symbols_path parameter, which only applies when the session is
first created. See open_windbg_dump and
--symbols-path for the details.
Get a structured report with the dump-triage prompt¶
For a thorough, consistent write-up, use the built-in dump-triage prompt instead of asking
free-form. Most clients surface it as a slash command or a prompt picker; in VS Code it appears
as /mcp.mcp-windbg.dump-triage. It walks the model through opening the dump, extracting
metadata, and producing a structured crash report. See Prompts.
Close the session when done¶
Each open dump holds a cdb.exe process. Free it when you finish:
Large dumps and timeouts
Heavy commands on big dumps can exceed the default 30-second command timeout. Raise it
with --timeout, for example --timeout 120.
Related¶
- Triage multiple dumps - run this flow across a whole folder.
- Tools reference - parameters for every tool.
- Troubleshooting - missing symbols, CDB not found, timeouts.