mcp server crashes when you cancel a long-running tool call mid-execution

tested the filesystem mcp server and github mcp server. both crash if you cancel a tool call that's taking more than ~10 seconds to execute. like if you call a tool that's reading a huge directory or cloning a large repo, and you ctrl+c or timeout the request, the mcp server process just dies instead of handling the cancellation gracefully tested on mcp sdk 0.9.1 and 0.9.2, same behaviour. the server logs 'SIGTERM received' and then segfaults. no cleanup, no error response back to the client, just dead has anyone built mcp servers that handle cancellation properly or is this just a known limitation of the sdk right now