StdioClientTransport missing explicit UTF-8 charset in InputStreamReader (same issue as #295, but on client side)

Comece em mcp-core/src/main/java/io/modelcontextprotocol/client/transport/StdioClientTransport.java, lendo startErrorProcessing e startInboundProcessing juntamente com o tratamento existente de UTF-8 em startOutboundProcessing. Compare a correção relacionada no lado do servidor em #826 e verifique se as respostas e os erros continuam sendo decodificados corretamente quando o conjunto de caracteres padrão da JVM não é UTF-8.

Bug description

StdioClientTransport has the same encoding mismatch issue that was identified in #295 and fixed for StdioServerTransportProvider in #826 — but the fix was only applied to the server side. The client transport still lacks explicit UTF-8 charset
specification when reading from the subprocess.

In startInboundProcessing, the InputStreamReader is created without specifying a charset:

https://github.com/modelcontextprotocol/java-sdk/blob/main/mcp-core/src/main/java/io/modelcontextprotocol/client/transport/StdioClientTransport.java#L249

try (BufferedReader processReader = new BufferedReader(new InputStreamReader(process.getInputStream()))) {

Similarly, in startErrorProcessing:

https://github.com/modelcontextprotocol/java-sdk/blob/main/mcp-core/src/main/java/io/modelcontextprotocol/client/transport/StdioClientTransport.java#L182-L183

try (BufferedReader processErrorReader = new BufferedReader(
        new InputStreamReader(process.getErrorStream()))) {

Meanwhile, startOutboundProcessing already correctly specifies UTF-8:

os.write(jsonMessage.getBytes(StandardCharsets.UTF_8));
os.write("\n".getBytes(StandardCharsets.UTF_8));

This is the exact same inconsistency that #295 reported for StdioServerTransportProvider, and that #826 fixed — only on the server side.

Steps to reproduce

1. Start a JVM with default charset set to something other than UTF-8 (e.g., -Dfile.encoding=COMPAT on Windows with Japanese locale, which resolves to MS932/Shift_JIS)
2. Connect to an MCP server via StdioClientTransport
3. Call a tool that returns multi-byte UTF-8 characters (e.g., Japanese, Chinese, Korean, emoji) in its response

Expected behavior

Multi-byte characters in the server's JSON-RPC response should be decoded correctly, since the MCP stdio transport specification requires UTF-8.

Actual behavior

The InputStreamReader uses Charset.defaultCharset() instead of UTF-8. When the default charset is not UTF-8, the response bytes are decoded with the wrong charset, corrupting multi-byte characters. This corruption can also break the JSON structure itself,
resulting in JsonParseException:

com.fasterxml.jackson.core.JsonParseException: Unexpected character ('' (code 92)): was expecting double-quote to start field name

For example, with MS932 as the default charset, the last byte of certain UTF-8 characters (0x8B, etc.) is interpreted as a MS932 lead byte, which then consumes the following byte — potentially a JSON structural character like \ (0x5C). This shifts the parser
state and breaks JSON parsing entirely.

Environment

• MCP Java SDK version: 1.1.1
• Java version: 21
• OS: Windows 11 (Japanese locale, default charset MS932 with -Dfile.encoding=COMPAT)