Problem

The stdio server transport wraps sys.stdin.buffer and sys.stdout.buffer with TextIOWrapper for UTF-8 encoding. TextIOWrapper calls close() in __del__, which closes the underlying buffer. When the server exits and the AsyncFile wrappers are garbage collected, this closes the real process stdio file descriptors (fd 0 and fd 1).

This causes ValueError on subsequent print() or input() calls in the parent process after the MCP server exits (e.g., Ctrl+D exit).

Reproduction

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("Demo")
mcp.run(transport="stdio")
print("?")  # Raises ValueError after Ctrl+D

Fix

Introduce _NoCloseTextIOWrapper that overrides close() and __del__() to prevent closing the underlying buffer. The standard process handles should outlive the server.

The comment in the existing code already states the intent: "Purposely not using context managers for these, as we don't want to close standard process handles." However, TextIOWrapper.__del__ calls close() regardless, which silently closes the buffer on garbage collection.

Testing

Verified that:

• _NoCloseTextIOWrapper does not close the underlying buffer when garbage collected
• sys.stdin remains readable and sys.stdout remains writable after the wrapper is deleted
• Standard TextIOWrapper does close the buffer (confirming the bug)

Fixes #1933