plugins/audio/mcp.py

from app.mcp.decorators import mcp_tool
from pydantic import BaseModel, Field

from .core import get_transcription_service


class TranscriptionOutput(BaseModel):
    success: bool = Field(description="是否成功")
    text: str = Field(description="转录文本")
    segments: list = Field(default_factory=list, description="分段结果")
    error: str | None = Field(default=None, description="错误信息")


@mcp_tool(
    name="audio-transcribe",
    title="音频转录",
    description="将 Base64 音频转录为文本",
    annotations={"readOnlyHint": True, "destructiveHint": False},
)
async def transcribe_audio_base64(
    audio_base64: str,
    suffix: str = ".wav",
    language: str | None = None,
) -> TranscriptionOutput:
    result = get_transcription_service().transcribe_base64(
        audio_base64,
        suffix,
        language,
    )
    return TranscriptionOutput(**result)


@mcp_tool(
    name="audio-file",
    title="音频文件转录",
    description="将本地音频文件转录为文本",
    annotations={"readOnlyHint": True, "destructiveHint": False},
)
async def transcribe_audio_file(
    audio_path: str,
    language: str | None = None,
) -> TranscriptionOutput:
    result = get_transcription_service().transcribe_file(audio_path, language)
    return TranscriptionOutput(**result)