中文 | English | 日本語 | Français | Deutsch

A high-performance offline translation model server with minimal resource requirements - no GPU needed. Average response time of 50ms per request. Supports translation of major languages worldwide.

Note: This model server focuses on offline translation, response speed, cross-platform deployment, and local execution to achieve unlimited free translation. Due to model size and optimization constraints, the translation quality will not match that of large language models. For high-quality translation, consider using online large language model APIs.

v4 has optimized memory usage, further improved speed, and enhanced stability. If you are using an old version, it is recommended to upgrade immediately!

Website	TOKEN	Other Interface	Provider
ipacel.cc	__IpacEL_MT_API_TOKEN__	Immersive Translate: https://MTranServer.ipacel.cc/imme?token=__IpacEL_MT_API_TOKEN__	@ApliNi

Thanks to community contributors for providing trial services for users!

Now supports desktop one-click launch! Supports Windows, Mac, Linux.

Download the latest desktop version for your platform from Releases, install and launch directly.

After the desktop app launches, it will create a tray menu to conveniently manage the service.

The program includes a simple UI and online debug documentation.

For detailed usage instructions, jump to Ecosystem Projects

Preview (latest version has updates):

It is recommended to use the desktop app or Docker deployment for better performance and convenience. Manual server deployment is for advanced users.

Programmers can start the server directly via command line:

npx mtranserver@latest

npx can be replaced with any package manager you prefer, such as bunx, pnpx, etc.

Important Note:

When translating a language pair for the first time, the server will automatically download the corresponding translation model (unless offline mode is enabled). This process may take some time depending on your network speed and model size. After the model is downloaded, subsequent translation requests will enjoy millisecond-level response speeds. It is recommended to test a translation once before formal use to allow the server to pre-download and load the model. The program is frequently updated. If you encounter problems, please try updating to the latest version.

npm i -g mtranserver@latest

npm can be replaced with any package manager you prefer, such as bun, pnpm, etc.

Then run mtranserver.

Create a compose.yml file in an empty directory, with the following content:

services:
  mtranserver:
    image: xxnuo/mtranserver:latest
    container_name: mtranserver
    restart: unless-stopped
    ports:
      - "8989:8989"
    environment:
      - MT_HOST=0.0.0.0
      - MT_PORT=8989
      - MT_OFFLINE=false
      # - MT_API_TOKEN=your_secret_token_here
    volumes:
      - ./models:/app/models

docker pull xxnuo/mtranserver:latest
docker compose up -d

Supports VS Code, Cursor, Augment, and other VS Code-based IDEs.

Search for MTranCode in the extension marketplace to install the comment translation plugin.

The plugin defaults to calling the server at http://localhost:8989 for comment and code translation. You can adjust it in settings.

This plugin is forked from vscode-comment-translate.

TODO: Under active development.

If you develop a derivative project, feel free to submit a PR. I will add your project to the ecosystem list.

By the way, the project has been published to npm. You can directly call the simple library interface in other programs to implement translation functionality. For specific information, check the TypeScript type definitions.

The server provides compatible endpoints for multiple translation plugins:

Endpoint	Method	Description	Supported Plugins
/imme	POST	Immersive Translation plugin endpoint	Immersive Translation
/kiss	POST	Kiss Translator plugin endpoint	Kiss Translator
/deepl	POST	DeepL API v2 compatible endpoint	Clients supporting DeepL API
/deeplx	POST	DeepLX compatible endpoint	Clients supporting DeepLX API
/hcfy	POST	Selection Translator compatible endpoint	Selection Translator
/hcfy	POST	Selection Translator compatible endpoint	Selection Translator
/google/language/translate/v2	POST	Google Translate API v2 compatible endpoint	Clients supporting Google Translate API
/google/translate_a/single	GET	Google translate_a/single compatible endpoint	Clients supporting Google web translation

Plugin Configuration Guide:

Note:

• Immersive Translation - Enable Beta features in developer mode in Settings to see Custom API Settings under Translation Services (official tutorial with images). Then increase the Maximum Requests per Second in Custom API Settings to fully utilize server performance. I set Maximum Requests per Second to 512 and Maximum Paragraphs per Request to 1. You can adjust based on your server hardware.

• Kiss Translator - Scroll down in Settings page to find the custom interface Custom. Similarly, set Maximum Concurrent Requests and Request Interval Time to fully utilize server performance. I set Maximum Concurrent Requests to 100 and Request Interval Time to 1. You can adjust based on your server configuration.

Configure the plugin's custom interface address according to the table below.

Name	URL	Plugin Setting
Immersive Translation (No Password)	http://localhost:8989/imme	Custom API Settings - API URL
Immersive Translation (With Password)	http://localhost:8989/imme?token=your_token	Same as above, change your_token to your MT_API_TOKEN value
Kiss Translator (No Password)	http://localhost:8989/kiss	Interface Settings - Custom - URL
Kiss Translator (With Password)	http://localhost:8989/kiss	Same as above, fill KEY with your_token
DeepL Compatible	http://localhost:8989/deepl	Use DeepL-Auth-Key or Bearer authentication
DeepLX Compatible	http://localhost:8989/deeplx	Support token parameter or Bearer authentication
Google Compatible	http://localhost:8989/google/language/translate/v2	Use key parameter or Bearer authentication
Selection Translator	http://localhost:8989/hcfy	Support token parameter or Bearer authentication

Regular users can start using the service after setting up the plugin interface address according to the table above.

./mtranserver [options]

Options:
  -version, -v          Show version information
  -log-level string     Log level (debug, info, warn, error) (default "warn")
  -config-dir string    Configuration directory (default "~/.config/mtran/server")
  -model-dir string     Model directory (default "~/.config/mtran/models")
  -host string          Server host address (default "0.0.0.0")
  -port string          Server port (default "8989")
  -ui                   Enable Web UI (default true)
  -offline              Enable offline mode, disable automatic model download (default false)
  -worker-idle-timeout int  Worker idle timeout in seconds (default 300)
  --download pairs...   Download models for specific language pairs (e.g. --download en_zh zh_en)
  --languages           List all supported language pairs for download

Note: `--download` and `--languages` require network access and do not work in offline mode.

Examples:
  ./mtranserver --host 127.0.0.1 --port 8080
  ./mtranserver --ui --offline
  ./mtranserver -v

Here are some similar projects, you can try them if you have other needs:

Project Name	Memory Usage	Concurrency	Translation Quality	Speed	Additional Info
NLLB	Very High	Poor	Average	Slow	Android port RTranslator has many optimizations, but still has high resource usage and is not fast
LibreTranslate	Very High	Average	Average	Medium	Mid-range CPU processes 3 sentences/s, high-end CPU processes 15-20 sentences/s. Details
OPUS-MT	High	Average	Below Average	Fast	Performance Tests
Any LLM	Extremely High	Dynamic	Very Good	Very Slow	High hardware requirements. If you need high concurrency translation, it is recommended to use vllm framework.
MTranServer (This Project)	Low	High	Average	Ultra Fast	50ms average response time per request.

Table data is for CPU, English to Chinese scenarios simple testing, not strict testing, non-quantized version comparison, for reference only.

Please refer to API_en.md and the API documentation after startup.

Bergamot Project for awesome idea of local translation.

Mozilla for the models.