Daisys MCP Server

import os from daisys import DaisysAPI # type: ignore # from daisys.v1.speak.models import ProsodyFeaturesUnion, ProsodyType from mcp.server.fastmcp import FastMCP # type: ignore from mcp.types import TextContent from typing import Literal from daisys_mcp.model import McpVoice, McpModel, VoiceGender from daisys_mcp.websocket_tts import text_to_speech_websocket from daisys_mcp.http_tts import text_to_speech_http from daisys_mcp.utils import throw_mcp_error, make_output_file, make_output_path from dotenv import load_dotenv # type: ignore load_dotenv() # Initialize FastMCP server mcp = FastMCP("Daisys-mcp-server") email = os.environ.get("DAISYS_EMAIL") password = os.environ.get("DAISYS_PASSWORD") if not email or not password: throw_mcp_error("DAISYS_EMAIL, DAISYS_PASSWORD environment variable is required") storage_path = os.environ.get("DAISYS_BASE_STORAGE_PATH") @mcp.tool( "text_to_speech", description=( """ Convert text to speech with a given voice and play the audio buffer. Voice_id can be provided. If not provided, the latest voice will be used. ⚠️ TOKEN WARNING: This tool makes an API call to Daisys API which may incur costs. Args: text (str): The text to convert to speech. voice_id (str, optional): The voice_id of the voice to use. If no voice specified use latest created voice. audio_format (str, optional): Can be either "wav" or "mp3". Defaults to "wav" always use "wav" unless mp3 specified. output_dir (str, optional): Directory where files should be saved. Defaults to $HOME/Desktop if not provided. streaming (bool, optional): Whether to use streaming or not. Set to True unless specifically asked to not stream. (streaming makes use of the websocket protocol which send and play audio in chunks) Defaults don't store if not provided. Returns: Text content with the path to the output file and name of the voice used. """ ), ) # Disabled optional typing since its not yet supported by cursor's mcp client def text_to_speech( text: str, voice_id: str = None, # type: ignore audio_format: str = "wav", output_dir: str = None, # type: ignore streaming: bool = True, ): if text in ["None", "", None]: throw_mcp_error("Text for TTS cannot be empty.") # LLM sometimes send null as a string if isinstance(voice_id, str) and voice_id.lower() in ["null", "undefined"]: voice_id = None # type: ignore with DaisysAPI("speak", email=email, password=password) as speak: # type: ignore if not voice_id: try: voice_id = speak.get_voices()[-1].voice_id except IndexError: throw_mcp_error("No voices available. Try to generate a voice first.") try: # this can create only a wav file but has fast inference if audio_format == "wav" and streaming: audiobuffer = text_to_speech_websocket(text, voice_id) else: audiobuffer = text_to_speech_http(text, voice_id) except Exception: throw_mcp_error("Error generating audio") if not storage_path: return TextContent( type="text", text=f"Success. Voice used: {voice_id}", ) # Create the output file output_path = make_output_path(output_dir, storage_path) output_file_name = make_output_file(text, output_path, audio_format) output_path.parent.mkdir(parents=True, exist_ok=True) with open(output_path / output_file_name, "wb") as f: f.write(audiobuffer) # type: ignore return TextContent( type="text", text=f"Success. File saved as: {output_path / output_file_name}. Voice used: {voice_id}", ) @mcp.tool( "get_voices", description="Get available voices can be filtered by model and gender, and sorted by name or timestamp in ascending or descending order.", ) # Disabled optional typing since its not yet supported by cursor's mcp client def get_voices( model: str = None, gender: str = None, sort_by: Literal["description", "name"] = "name", sort_direction: Literal["asc", "desc"] = "asc", ): with DaisysAPI("speak", email=email, password=password) as speak: filtered_voices = [ voice for voice in speak.get_voices() if (model is None or voice.model == model) and (gender is None or voice.gender == gender) ] voice_list = [ McpVoice( voice_id=voice.voice_id, name=voice.name, gender=voice.gender, model=voice.model, description=voice.description, ) for voice in filtered_voices ] if sort_direction == "asc": voice_list.sort(key=lambda x: getattr(x, sort_by).lower()) else: voice_list.sort(key=lambda x: getattr(x, sort_by).lower(), reverse=True) return voice_list @mcp.tool( "get_models", description=( """ Get all available models from Daisys API. Args: language (str, optional): needs to be "de" for german, "en" for english and "nl" for dutch. Defaults to None. sort_by (str, optional): can be "name" or "displayname". Defaults to "displayname". sort_direction (str, optional): can be "asc" or "desc". Defaults to "asc". Returns: model_list: An object containing details of all models """ ), ) # Disabled optional typing since its not yet supported by cursor's mcp client def get_models( language: str = None, sort_by: Literal["name", "displayname"] = "displayname", sort_direction: Literal["asc", "desc"] = "asc", ): # make sure to only use the first 2 letters of the language if language: language = language.lower()[:1] with DaisysAPI("speak", email=email, password=password) as speak: filtered_models = [ model for model in speak.get_models() if language is None or any(lang.startswith(language) for lang in model.languages) ] model_list = [ McpModel( name=model.name, displayname=model.displayname, flags=model.flags, languages=model.languages, genders=model.genders, styles=model.styles, prosody_types=model.prosody_types, ) for model in filtered_models ] if sort_direction == "asc": model_list.sort(key=lambda x: getattr(x, sort_by).lower()) else: model_list.sort(key=lambda x: getattr(x, sort_by).lower(), reverse=True) return model_list @mcp.tool( "create_voice", description=( """ Convert text to speech with a given voice and play the audio buffer. Voice_id can be provided. If not provided, the latest voice will be used. before calling also call get models so it always inputs a valid model ⚠️ TOKEN WARNING: This tool makes an API call to Daisys API which may incur costs. Args: name (str, optional): The name of the voice to create. Defaults to "Daisy". gender (str, optional): The gender of the voice can be "male" or "female". Defaults to "female". model (str, optional): The model of the voice. Defaults to "english-v3.0". pitch (int, optional): Adjusts the pitch level; -10 for very low pitch, 10 for very high pitch. pace (int, optional): Controls the speech rate; -10 for very slow, 10 for very fast. expression (int, optional): Modulates expressiveness; -10 for monotone, 10 for highly expressive. Returns: McpVoice: An object containing details of the created voice, including voice_id, name, gender, model, and description. """ ), ) def create_voice( name: str = "Daisy", gender: str = "female", model: str = "english-v3.0", pitch: int = 0, pace: int = 0, expression: int = 0, ): # Using SimpleProsody prosody_params = { "pitch": pitch, "pace": pace, "expression": expression, } # Validate that all prosody parameters are within the range -10 to 10 for param_name, value in prosody_params.items(): if not -10 <= value <= 10: throw_mcp_error(f"{param_name} must be between -10 and 10, got {value}.") if gender not in VoiceGender: throw_mcp_error( f"Invalid gender: {gender}. Must be one of {list(VoiceGender)}." ) with DaisysAPI("speak", email=email, password=password) as speak: voice = speak.generate_voice( name=name, gender=gender, model=model, default_prosody=prosody_params ) return McpVoice( voice_id=voice.voice_id, name=voice.name, gender=voice.gender, model=voice.model, description=voice.description, ) @mcp.tool( "remove_voice", description="Delete a voice.", ) def remove_voice( voice_id: str, ): with DaisysAPI("speak", email=email, password=password) as speak: speak.delete_voice(voice_id) return TextContent( type="text", text=f"Success. voice {voice_id} deleted.", ) def main(): print("Starting Daisys-mcp server.") mcp.run(transport="stdio") if __name__ == "__main__": main()