| Issue |
169703
|
| Summary |
`lldb-mcp` will hang on interactive commands
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
AKil47
|
# The Issue
When a command like `breakpoint delete` is sent via MCP to `lldb`, the `lldb` session will get an interactive prompt that the user needs to confirm.
For example:
```
lldb) breakpoint delete
About to delete all breakpoints, do you want to do that?: [Y/n]
```
The problem is that the MCP server is not made aware that it needs to wait, and thus, if the user is not actively monitoring the `lldb` terminal, they will not realize that the command is hanging because it is waiting for acknowledgement, resulting in timeouts.
For example in Codex:
```
• Called lldb-mcp.lldb_command({"debugger_id":1,"arguments":"br delete"})
└ Error: tool call error: tool call failed for `lldb-mcp/lldb_command`
Caused by:
0: timed out awaiting tools/call after 60s
1: deadline has elapsed
```
# Proposed Solution
I think the easiest would be to have something like `About to delete all breakpoints, do you want to do that?: [Y/n]` be the response that is sent over MCP, so that the LLM recognizes it is being asked a question. It can then send `Y` or `n` back as a response (after consulting users)
Since some users might not _want_ an interactive command to be exposed over MCP, it might make sense to add some type of flag to enable/disable this command.
Happy to try to take a look at this myself is the maintainers consider it a real issue!
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
