Synthesize text into speech and play audio via the evo tts CLI or MCP server.
npx skills add https://github.com/maycuatroi1/agent-skills --skill speak-maycuatroi1
Please help me install this Agent Skill. Skill: speak Source: https://github.com/maycuatroi1/agent-skills/tree/main/skills/speak Command: npx skills add https://github.com/maycuatroi1/agent-skills --skill speak-maycuatroi1
evo tts CLI and an evo-tts MCP server so an agent can hand over results by voice instead of a wall of text. ## Core Features & Use Cases - Two synthesis providers: Vbee for natural Vietnamese with correct tones, and OpenAI gpt-4o-mini-tts for English and steerable delivery via instructions, with automatic provider selection based on available credentials. - Realtime and batch modes: evo tts speak synthesizes and plays immediately with sentence-boundary chunking for long text, while evo tts batch processes directories of files or JSONL manifests through the async API. - MCP server integration: A stdio MCP server exposes speak, speak_batch, and list_voices tools, registered with Claude Code, Codex, and OpenCode via one idempotent install script. - Use Case: After finishing a long build or deployment, have the agent speak a short Vietnamese or English summary through your speakers so you know the result without reading the terminal. ## Quick Start Ask the agent to speak a short summary of the finished task out loud, for example: "Say it out loud: the build finished and all tests passed."High-intent search queries and answers about installing and using this skill.
Run `evo tts speak "your text"` to synthesize and play audio immediately. You can also pipe stdin, read a file with `-f`, save output with `-o`, or stream raw bytes with `--stdout`.
Vbee provides natural Vietnamese voices with correct tones via its API, and is the default provider when its credentials exist. OpenAI gpt-4o-mini-tts handles English and other languages with steerable delivery through instructions.
Run `bash mcp/tts/install.sh` to register the evo-tts stdio MCP server with every agent CLI found on PATH. It exposes speak, speak_batch, and list_voices tools and is idempotent.
A 401 means the Vbee token expired according to the app's setting of 7, 30, 60, or 90 days. Create a new app at studio.vbee.vn and re-add the app ID and token with `evo cred add`.
Vbee realtime accepts 300 characters per request and batch accepts 100,000, while OpenAI allows 4,000 characters per request with no batch endpoint. Longer realtime text is split on sentence boundaries and the audio chunks are joined.
Playback requires ffplay, mpv, cvlc, afplay, paplay, or aplay on the system. If none is found the CLI warns but still writes the audio file; installing ffmpeg fixes playback on every platform.
AlexKapadiaFetch Kaggle datasets into a Pixie tool's data folder with credentials.
sakamotomomotaro0809-netizenConvert Japanese text into speech audio using macOS say, Google Cloud TTS, VOICEVOX, or OpenAI TTS.
pulzzeManage OAuth tokens and API keys with automatic refresh and namespace isolation.
skills-ilBuild Hebrew voice bots and IVR systems with telephony integration.