bug(server): text_to_audio validates text outside try/except, inconsistent with other tools

Summary

In minimax_mcp/server.py, the input validation for text_to_audio is performed outside the try/except MinimaxAPIError block, while every other tool function performs its validation inside the block. This means invalid input to text_to_audio propagates as an unhandled MinimaxRequestError exception, while invalid input to peer tools is caught and returned as a TextContent error message.

Code shape

text_to_audio (around line 90-91):

def text_to_audio(
    text: str, ...
) -> TextContent:
    if not text:
        raise MinimaxRequestError("Text is required.")  # <-- OUTSIDE try/except

    try:
        # ... main work ...
    except MinimaxAPIError as e:
        return TextContent(type="text", text=f"Failed to convert text to audio: {str(e)}")

text_to_image, music_generation, voice_clone, etc. all do the equivalent check inside the try block, so their MinimaxRequestError is caught and returned as a TextContent.

Impact

• Inconsistent client experience: a missing text to text_to_audio crashes the tool call; the same kind of missing input to text_to_image returns a friendly error.
• An MCP client that doesn't catch MinimaxRequestError (and most don't, since it's an implementation detail) sees a tool crash instead of a structured error.

Suggested fix

Move the if not text: raise MinimaxRequestError(...) check from outside the try to inside it, matching the pattern in the other tools. This is a one-line move.

Discovered via

Issue filed as a follow-up to PR #87 (test coverage). The test added in #87 (test_text_to_audio_with_empty_text_raises) currently expects MinimaxRequestError to propagate, but to make the behavior consistent with the rest of the file, that expectation should probably flip to a TextContent return value.

🤖 Generated with Claude Code

Start in minimax_mcp/server.py at text_to_audio and compare its validation and try/except structure with text_to_image, music_generation, and voice_clone. Run test_text_to_audio_with_empty_text_raises from PR #87, then verify that empty text produces the same structured TextContent error behavior as the other tools.