LLDB server accepts only 4 clients; 5th connection hangs

I’m seeing an issue where the LLDB MCP server appears to accept only 4 client connections. On the 5th run, the server stops responding and the client times out.

How to reproudce

I run the same client sample repeatedly (sequentially).
Each run executes the close/cleanup logic (I have confirmed the close path is reached).
After the 4th successful run, the 5th connection attempt does not get a response from the server, and the client eventually times out.

My Question

Could you please help confirm whether this is:

a limitation/bug in the LLDB server (e.g., max client connections = 4, sockets not being released, etc.), or
an issue in my client close flow (e.g., connection not fully closed, resources leaked, etc.)?

BTW, the log folder /tmp/diagnostics-6c50da is None

New MCP client connected: client_4
LLDB diagnostics will be written to /tmp/diagnostics-6c50da
Please include the directory content when filing a bug report
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.

    def __init__(self, host: str = "localhost", port: int = 8085, timeout: float = 120.0,
                 verbose: bool = False):
        self.host = host
        self.port = port
        self.timeout = timeout
        self.verbose = verbose
        self._sock: Optional[socket.socket] = None
        self._request_id = 0
        self._recv_buf = b""

    def connect(self):
        self._sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        self._sock.settimeout(self.timeout)
        self._sock.connect((self.host, self.port))
        self._recv_buf = b""

    def close(self):
        if self._sock:
            try:
                self._sock.close()
            except OSError:
                pass
            self._sock = None
            self._recv_buf = b""

Start by reproducing the sequential client runs against the LLDB MCP server on port 8085 and observe the fifth connection. Inspect the server-side connection cleanup alongside the client’s close() path, then check whether /tmp/diagnostics-6c50da is populated. Done means identifying whether the hang is caused by server-side connection handling or incomplete client cleanup and documenting the evidence.