Skip to content

Let an AI Agent Speak: Text to Speech (TTS) over MCP and UTCP

Connect an AI agent to the OVOS text-to-speech (TTS) server, so the agent can turn text into a WAV file. Use the TTS server when Claude Code, Claude Desktop, opencode, Mistral Vibe, or your own MCP or UTCP agent must make speech with a TTS voice that runs on your own computer.

What you get

  • An MCP tool synthesize at http://localhost:9666/mcp. It returns the WAV audio as base64.
  • A UTCP manual at http://localhost:9666/utcp.

Start the server

Alpha channel only

The mcp extra needs the prerelease channel:

pip install --pre "ovos-tts-server[mcp]" ovos-tts-plugin-espeakng

  1. Start the server with the --mcp flag:

    ovos-tts-server --engine ovos-tts-plugin-espeakng --mcp

Speak from Claude Code

  1. Add the server to Claude Code:

    claude mcp add --transport http ovos-tts http://localhost:9666/mcp

  2. Make sure that Claude Code can connect:

    claude mcp list

    ovos-tts: http://localhost:9666/mcp (HTTP) - ✔ Connected

Speak from Claude Desktop

Add the mcp-remote bridge to claude_desktop_config.json, then restart Claude Desktop. The bridge needs Node.js.

{
  "mcpServers": {
    "ovos-tts": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://localhost:9666/mcp"]
    }
  }
}

Speak from opencode

  1. Add the server to opencode.json in your project folder:

    {
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "ovos-tts": {
      "type": "remote",
      "url": "http://localhost:9666/mcp",
      "enabled": true
    }
  }
}

  2. Make sure that opencode can connect:

    opencode mcp list

    ●  ✓ ovos-tts connected
│      http://localhost:9666/mcp

Speak from Mistral Vibe

  1. Add the server to ~/.vibe/config.toml:

    [[mcp_servers]]
name = "ovos-tts"
transport = "streamable-http"
url = "http://localhost:9666/mcp"

    Write the entry in the file. Do not use vibe mcp add for this server: that command sets OAuth login for an HTTP server, and the OVOS server does not use OAuth.

  2. Start vibe. Vibe loads the tools of the server when it starts. It gives each tool the server name as a prefix and changes each hyphen to _, for example ovos_tts_synthesize.

Speak from your own MCP client

This script calls synthesize and writes the audio to speech.wav. It uses the mcp SDK (pip install "mcp>=2.3.0") in any environment. The client does not need the environment of the server.

import asyncio, base64, json
from mcp import ClientSession
from mcp.client.streamable_http import streamable_http_client

async def main():
    async with streamable_http_client("http://localhost:9666/mcp") as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            result = await session.call_tool(
                "synthesize", {"text": "turn on the kitchen light", "lang": "en-us"})
            reply = json.loads(result.content[0].text)
            open("speech.wav", "wb").write(base64.b64decode(reply["data"]))
            print(reply["mime_type"], len(reply["data"]), "base64 characters")

asyncio.run(main())

audio/wav 88896 base64 characters

file speech.wav reports RIFF (little-endian) data, WAVE audio, Microsoft PCM, 16 bit, mono 22050 Hz. To transcribe this file, see Transcribe audio from an AI agent.

Speak from a UTCP agent

The UTCP manual lists tts_status and tts_synthesize_v2. tts_status returns the plugin and its languages. tts_synthesize_v2 returns the same JSON object as the MCP tool synthesize:

import asyncio, base64
from utcp.utcp_client import UtcpClient

async def main():
    client = await UtcpClient.create(config={"manual_call_templates": [{
        "name": "ovos_tts", "call_template_type": "http",
        "url": "http://localhost:9666/utcp", "http_method": "GET"}]})
    print(await client.call_tool("ovos_tts.tts_status", {}))
    reply = await client.call_tool("ovos_tts.tts_synthesize_v2",
                                   {"utterance": "turn on the kitchen light", "lang": "en-us"})
    open("speech.wav", "wb").write(base64.b64decode(reply["data"]))
    print(reply["mime_type"])

asyncio.run(main())

{'status': 'ok', 'plugin': 'ovos-tts-plugin-espeakng', 'langs': ['fr-ch', 'th', 'uz', 'en-us', ...], 'default_lang': 'en-us', 'default_model': None, 'default_voice': 'm1'}
audio/wav

Tool reference

Tool Protocol Arguments Result
synthesize MCP text, voice (optional), lang (optional) {"mime_type": "audio/wav", "data": <base64 WAV>, "path": <file on the server>, "phonemes"}
tts_status UTCP none {"status", "plugin", "langs", ...}
tts_synthesize_v2 UTCP utterance, voice (optional), lang (optional) {"mime_type", "data", "phonemes"}

If it fails

What you see Cause What to do
/mcp returns 404 The server started without --mcp, or the mcp extra is not installed Install the mcp extra. Start the server with --mcp. The server log shows a warning when the extra is missing.
A client behind a TLS reverse proxy refuses a redirect The server answers POST /mcp with 307 and the location /mcp/ Use http://<host>:9666/mcp/, with the slash at the end.

More information

Read next: Translate text from an AI agent Related: AI Agents · Transcribe audio from an AI agent · TTS Server