Voice MCP

What forgemeshlabs/voice-mcp MCP server does

The forgemeshlabs/voice-mcp MCP server gives an MCP client access to speech generation through https://voice.forgemesh.io. It can turn text into WAV audio using 10 standard voices, 10 persona voices, and 31 supported language codes. Generated audio is returned as audio_base64, allowing the calling client or application to store, play, or forward the WAV bytes.

A free catalog tool exposes the available voices, persona names, languages, pricing buckets, character limits, and supported controls before an agent makes a paid request. The paid tools cover low-cost standard speech, controlled speech with speed and quality settings, expressive persona speech, OpenAI-shaped requests, and batch generation.

How it works

The forgemeshlabs/voice-mcp MCP server validates request values locally before sending a paid x402 call. Validation includes voice names, language codes, audio formats, speed and quality ranges, batch size, and text length. The available speed range is 0.7x to 2.0x, while quality uses steps from 1 through 100.

Payments are made per call in USDC on Base. The server requires access to a Base wallet through its private key for paid tools. No voice-service API key or subscription is required. Standard and OpenAI-compatible generation cost $0.001 for 1–500 characters and $0.003 for 501–2,000 characters. Controlled generation costs $0.003 or $0.006 for those same length bands; persona generation costs $0.005 or $0.01; batch generation costs $0.002 or $0.005 based on total characters.

Setup and configuration

Install the npm package globally with:

npm install -g @forgemeshlabs/voice-mcp

A stdio MCP configuration can start the installed voice-mcp command and provide WALLET_PRIVATE_KEY in its environment. The key must belong to a Base wallet funded with USDC if paid tools will be used. Docker is also supported: build the repository image with the supplied Dockerfile and run it interactively with the wallet key passed as an environment variable.

Two optional settings are documented. X402_VOICE_BASE_URL changes the voice API base URL, and BASE_RPC_URL changes the Base RPC endpoint. The documented defaults are https://voice.forgemesh.io and https://mainnet.base.org respectively.

Tools and capabilities

• list_voice_catalog provides free capability and pricing discovery.
• generate_standard_voice creates speech with the 10 standard voices.
• generate_controlled_voice adds speed and quality controls.
• generate_persona_voice selects personas such as Storyteller, Narrator, Assistant, Urgent, Sage, or Echo.
• generate_openai_compatible_voice accepts an OpenAI-shaped request with input, voice, model, and response_format fields.
• generate_batch_voices processes up to 20 text items in one call using standard voices.

The forgemeshlabs/voice-mcp MCP server is suitable for agent narration, status messages, alerts, branded or character dialogue, scripted sequences, queues, and applications already structured around /v1/audio/speech-style payloads.

Limitations and notes

The catalog lookup is free, but speech generation is paid. Text length and batch limits apply, with the documented maximum of 2,000 characters for the supported single-text price bands and up to 20 items per batch. Audio tools return WAV data encoded in base64 rather than an audio file path or hosted download URL. A Base wallet private key and sufficient USDC are therefore needed for production use of paid capabilities.