MimikaStudio MCP
A local MCP server for voice work on macOS: voice cloning, speech synthesis and audiobooks from PDF on an agent's command
- 751 ★ on the repository
- Last commit
- GPL-3.0
Broad access
Why this level
- Cloning a voice from a three-second sample lets anyone fake a living person's speech. Without that person's consent it breaks the law and harms them directly: such recordings are used to deceive relatives and colleagues and to confirm payments and deals in someone else's name
- The `audiobook_generate_from_file` tool takes an absolute path, so the agent can reach any document your user can read
- The install script adds Homebrew packages, creates a Python environment and downloads large models from external sources
Install
- Works in
Manual install
git clone https://github.com/BoltzmannEntropy/MimikaStudio.git && cd MimikaStudio && ./install.shInstalls Homebrew packages, the Python environment, dependencies and the database. The first install takes a few minutes.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Cloning a voice from a three-second sample lets anyone fake a living person's speech. Without that person's consent it breaks the law and harms them directly: such recordings are used to deceive relatives and colleagues and to confirm payments and deals in someone else's name
- The `audiobook_generate_from_file` tool takes an absolute path, so the agent can reach any document your user can read
- The install script adds Homebrew packages, creates a Python environment and downloads large models from external sources
This is third-party code. Review the repository files before installing.
What it does
MimikaStudio runs a voice backend and an MCP server for it on your Mac. The server listens on port 8010, takes JSON-RPC calls over HTTP and gives an agent more than 50 tools: speech synthesis on the Kokoro, Qwen3-TTS and Chatterbox engines, voice management (list, upload, delete, preview), audiobook generation, system info, language model settings and an audio library. A voice can be cloned from a sample as short as three seconds. An audiobook is assembled from PDF, DOCX, EPUB, Markdown or TXT with a chapter queue and reusable voice presets. Heavy jobs go through a job queue, so the agent starts work and then polls its status. For instance, the `audiobook_generate_from_file` tool takes a document path, voice, speed and output format, and `audiobook_status` reports progress. All computation runs on the device through MLX on Apple Silicon, and no audio is uploaded anywhere. A Flutter window sits on top of the MCP and REST endpoints: the settings section lists every tool with parameters and shows service status. `./bin/mimikactl up` starts everything, and the `--no-flutter` flag skips the window.
Who it is for. For owners of Macs with Apple Silicon who want to voice texts and documents with local models and drive it from Claude Code or Codex.
Good fit when
- You need to turn a PDF with notes or an article into an MP3 audiobook without sending the file to a cloud
- An agent should voice a long text with a British Kokoro voice and watch the job to the end
- You want to clone your own voice from a short sample for narrating clips or notes
- A script has to call speech synthesis over HTTP on local port 8010 without a GUI
- You want to compare how the same text sounds in Kokoro, Qwen3-TTS and Chatterbox
Not a fit when
- You use Windows or Linux: ready builds exist only for macOS, and on other systems you would have to run the code yourself
- You want the voice of someone who has not consented: cloning without consent harms the voice owner and breaks the law
- You need narration live during a call or stream: jobs go through a queue, and the finished file appears in `backend/outputs/` after completion
- You need a cloud service with billing and support: MimikaStudio is a local app under the BSL license
Example request
Call Mimika MCP audiobook_generate_from_file for /Users/me/notes.pdf with voice bf_emma and format mp3, and report the finished file's address.MCP
- Transport
- http
- Authentication
- not required
Environment variables
SKIP_DICTA- A value of 1 skips downloading the Dicta Hebrew diacritizer model (about 1.1 GB) during install.
Limitations
`./install.sh` installs Homebrew packages, creates a Python virtual environment, pulls heavy dependencies and downloads the Dicta ONNX diacritizer of about 1.1 GB for Hebrew. You can skip that download with `SKIP_DICTA=1`. Speech models download on first run.
Besides the MCP server on port 8010, the app starts a backend API on port 7693. The MCP server asks for no key or login, so any process on your Mac can call its tools. The source is under the Business Source License 1.1 and binaries under a separate license, so the project is not freely licensed. The latest release is v2026.04.1, and the repository was last updated on 27 October 2026.
How to disable. Stop the MCP service with `./bin/mimikactl mcp stop`, or everything at once with `./bin/mimikactl down`. Remove the `http://127.0.0.1:8010` entry from the agent configuration.
FAQ
How do I start the MimikaStudio MCP server?
Run `./install.sh`, activate the environment with `source venv/bin/activate` and start `./bin/mimikactl up --no-flutter`. The server opens on port 8010, and `./bin/mimikactl mcp status` reports its state.
How do I connect MimikaStudio to Claude Code or Codex?
Add the local HTTP MCP server at `http://127.0.0.1:8010` in the client settings. Then ask the agent to call a tool such as `audiobook_generate_from_file` and track the job through `audiobook_status`.
Does MimikaStudio work offline?
After the models are downloaded, yes: computation runs on your Mac through MLX and audio does not leave for third-party servers. Models download on first run.
Which speech engines does MimikaStudio include?
Kokoro-82M for fast English synthesis, Qwen3-TTS in several sizes for cloning and preset voices, Chatterbox Multilingual with 23 languages for cloning, and Supertonic-2 on ONNX.
Does MimikaStudio run on Windows?
There are no ready Windows builds: binaries ship only for macOS on Apple Silicon. The code is cross-platform, but on other systems you run it yourself, and the author promises a Windows release later.
Discussion
No questions or reviews yet.
Related
MCP serversLocal speech synthesis for agents: a command and a skill voice text, clone a voice from a recording and render scores to audio, with no cloud
A text-to-speech MCP server that turns text and multi-voice scripts into audio through ElevenLabs and keeps a job history.
A local sound studio with an MCP server: an agent makes effects, music, songs and speech on open-weight models
A local MCP server generates music and sound effects, builds loops, adjusts loudness and exports audio for games or video.
Official MiniMax MCP server: text to speech, voice cloning and design, image and video generation
The agent speaks through your speakers, listens on the mic, translates across 15 languages and reads Thai PDFs via Paxa Labs
README in short
The author describes MimikaStudio as a local app with four capabilities: voice cloning, speech synthesis, reading documents aloud and audiobook creation. Audio samples from each engine, a table of models, `mimikactl` commands and a step-by-step PDF to audiobook flow over MCP complete the description.