VoiceStudio (debpalash) — Open-Source Local ElevenLabs Alternative · Agent Skill & MCP

What it is, and what it is not

• It is desktop, Docker and phone software that runs voice models on your hardware. The app asks before it downloads any model; installed models can run offline.
• It ships an agent skill (skills/voicestudio/SKILL.md) and mounts an MCP endpoint, so Claude Code, Codex, Cursor and other agents can generate speech and transcribe through your local backend.
• It is not a hosted cloud TTS service, and not something you can use from this page.

The skill is not the app

Installing the skill only teaches your agent how to talk to VoiceStudio. The agent still needs a running backend, by default http://localhost:3900. Start the desktop app or the Docker container first, then add the skill or the MCP connection.

Install

Commands copied from the upstream README and docs/install/docker.md at commit 4d4cb0badf99 on 31 Oct 2026. We have not run them ourselves; check the README before you install.

curl -fsSL https://voicestudio.sh/install | sh

irm https://voicestudio.sh/install.ps1 | iex

export OMNIVOICE_API_KEY="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"
docker pull ghcr.io/debpalash/voicestudio:stable

docker run -d --name omnivoice \
  -p 127.0.0.1:3900:3900 \
  -e OMNIVOICE_API_KEY="$OMNIVOICE_API_KEY" \
  -v omnivoice-data:/app/omnivoice_data \
  ghcr.io/debpalash/voicestudio:stable

npx skills add debpalash/VoiceStudio

Download installers from the latest GitHub release · Full Docker guide · Let a coding agent install it (docs/install/agent.md)

GPU or CPU?

No GPU is required. Upstream says the app is fully usable on the CPU, just slower; native GGUF engines can also use Vulkan on AMD, Intel and NVIDIA GPUs. We do not quote speeds; see the upstream performance page.

Connect over MCP

The running backend mounts an MCP endpoint. Point any Streamable HTTP client at it and keep the trailing slash:

http://localhost:3900/mcp/

The desktop app exports ready-made configs under Integrations for Claude Code (.mcp.json), Cursor (.cursor/mcp.json) and Codex CLI. The Codex CLI table from docs/mcp.md:

[mcp_servers.voicestudio]
url = "http://127.0.0.1:3900/mcp/"
http_headers = { "X-OmniVoice-Client-Id" = "codex-cli" }

Clients that only speak stdio use the bundled Python shim (backend.mcp_shim). Upstream warns that the MCP transport is not authenticated: keep it on localhost or a trusted LAN, never on the open internet. Read docs/mcp.md

Minimal agent flow

Steps from the upstream SKILL.md. We have not run this flow ourselves yet.

1. Check GET /health on http://localhost:3900 (or the address you configured).
2. Discover contracts with GET /openapi.json and voices with GET /v1/audio/voices. Do not invent profile IDs.
3. Generate speech with POST /v1/audio/speech, or transcribe with POST /v1/audio/transcriptions.
4. Check the HTTP status and probe the saved file before calling it audio; an error body can land at the output path.

Authorized voices and AGPL

• Clone a voice only with its owner’s permission. This is upstream’s rule and ours.
• AGPL-3.0 covers the app, including commercial use under its terms. If you modify it and offer it over a network, AGPL obligations apply. Model weights, voices and generated output have their own licences, which the Model Manager shows. LICENSE-NOTICE.md
• Listing is not an audit. AgentSkillsDir has not reviewed this code for security. Read install scripts before running them.

FAQ

Related

Browse all Claude skills in the directory

Corrections and suggestions

Spotted something out of date, or want to suggest a skill? Email a correction. To suggest a resource, submit it for manual review.