Character encoding of stdin/stdout is platform-dependent, leading to Unicode corruption

[rnortman]

Describe the bug
Character encoding of stdin/stdout on Windows is not UTF-8. (It will depend on the Windows locale, but usually CP437 or CP1252.) This breaks JSON, which requires proper Unicode. The Windows Claude Desktop application at least sense UTF-8 encoded data.

To Reproduce
Steps to reproduce the behavior:

1. Run Claude Desktop on Windows with an MCP server
2. Send something with an emoji
3. Observe corruption

Expected behavior
Unicode characters are not corrupted. (MCP clients/servers should be using UTF-8; see additional context below.)

Screenshots
N/A

Desktop (please complete the following information):

• OS: Windows 11
• Browser: None (Claude Desktop)
• Version: ???

Smartphone (please complete the following information):
N/A

Additional context

The JSON spec requires UTF-8, UTF-16, or UTF-32 encoding, so the default behavior on Windows is out of spec. The MCP protocol spec does not specify encoding and has no handshake to establish it, so this leaves some ambiguity, but I think UTF-8 is by far the most reasonable default in the modern era.

I have put up a PR with a fix: #198

[rnortman]

PR with fix: #198