Kokoro Text to Speech MCP server generates MP3 audio files from text using the Kokoro TTS model. It relies on Kokoro Onnx weights (kokoro-v1.0.onnx) and voices (voices-v1.0.bin) stored in the same repository and references the Kokoro-TTS source from HuggingFace Spaces. The server can optionally upload generated MP3s to S3, configurable via environment variables and the MCP config. The configuration example shows how to set the command, arguments, and environment variables such as TTS_VOICE, TTS_SPEED, TTS_LANGUAGE, AWS creds, AWS_REGION, AWS_S3_FOLDER, S3_ENABLED, and MP3_FOLDER. FFmpeg is required to convert WAV to MP3. Local MP3 files are stored in MP3_FOLDER, with optional automatic cleanup controlled by MP3_RETENTION_DAYS and DELETE_LOCAL_AFTER_S3_UPLOAD. S3 uploads can be toggled via S3_ENABLED and can be disabled per-request using the client’s --no-s3 option. The server runs with UV (uv run mcp-tts.py) and exposes a client tool (mcp_client.py) to submit TTS requests with options for text, file input, voice, and speed. Endpoint and bucket details, as well as host/port bindings, are configurable through environment variables.