fix(stdio_client): tolerate invalid UTF-8 from child stdout

[shaun0927]

Fixes #2454

Summary

stdio_client() currently decodes child stdout with encoding_error_handler="strict", so a malformed UTF-8 line can raise during TextReceiveStream(...) iteration and crash the transport task group.

This PR makes the client follow the same strategy that stdio_server() already uses after #2302:

• decode with errors="replace"
• let malformed lines fail JSON validation
• surface those failures as in-stream exceptions instead of killing the transport

It also treats abrupt child shutdowns (BrokenResourceError / ConnectionResetError) as normal shutdown-time conditions in the background stdio tasks.

Motivation and Context

PR #2302 fixed the server side so invalid UTF-8 on stdin becomes a parse error instead of a transport crash. The client side still behaved asymmetrically: a buggy or non-compliant child process could kill the Python client with a single malformed stdout line even if the next line was valid JSON-RPC.

That makes the client less robust than the server for the same class of malformed stdio input.

How Has This Been Tested?

Added a focused regression test in tests/client/test_stdio.py that spawns a child process which writes:

1. one malformed UTF-8 line (b"\xff\xfe\n")
2. one valid JSON-RPC line

Without this patch, the transport fails before the valid line is delivered.

With this patch:

• the first item from the read stream is an Exception
• the second item is the valid SessionMessage

Local verification:

• uv run --frozen pytest tests/client/test_stdio.py
• uv run --frozen ruff check src/mcp/client/stdio.py tests/client/test_stdio.py
• uv run --frozen pyright src/mcp/client/stdio.py

Breaking Changes

None.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

Checklist

• I have read the MCP Documentation
• My code follows the repository's style guidelines
• New and existing tests pass locally
• I have added appropriate error handling
• I have added or updated documentation as needed

[shaun0927]

for additional signal: mcp-claude[bot] independently reproduced this on both main and v1.x and arrived at the same one-line default change (issue #2454 comment). all 25 CI checks are green; happy to address review feedback.

[Bartok9]

Heads up @shaun0927 — this PR has gone conflicting after the stdio transport was refactored on main. I rebuilt the same root-cause fix (defaulting encoding_error_handler to "replace") cleanly against current main with a mid-session regression test in #2873, crediting your original approach here. Happy to defer if you'd rather refresh this one instead.

[maxisbey]

Thanks for the PR, and sorry it sat here without a proper review.

We're closing most of the open PR backlog. v2 is out and changed a lot of the SDK, so many older PRs no longer apply as written, and we're a small team that realistically doesn't have the capacity to work through the rest.

If this still matters to you on v2, the most useful thing you can do is open an issue (or comment on the existing one) with your use case and a repro. Hearing why it matters to you is what we use to decide what to prioritise.

_{AI Disclaimer}