分级翻译MCP服务

levelang.app MCP Server

An MCP server that exposes the levelang.app translation API to AI assistants. Unlike standard translators that always produce native-speaker complexity, levelang.app constrains translations to the learner's proficiency level.

Features

• Level-Aware Translation — Translate text at beginner, intermediate, advanced, or fluent proficiency with grammar constraints enforced per level
• Multi-Language Support — French, German, Italian, Mandarin Chinese, Cantonese, Japanese, with transliteration where applicable
• Mood Control — Casual and formal translation styles, plus honorific (keigo) for Japanese
• Language Discovery — Query available languages, levels, and moods dynamically from the backend
• MCP Resources — levelang://languages and levelang://languages/{code} for pulling language configs into context
• Stateless Wrapper — No database, no shared state; translates MCP tool calls into backend HTTP requests

Quick Start

You need two things: the MCP server URL and an API key. No local setup required.

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "levelang": {
      "url": "https://your-mcp-server-url/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "levelang": {
      "url": "https://your-mcp-server-url/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

Restart Claude Desktop. A hammer icon in the chat input indicates MCP tools are available.

Example Usage

Once connected, ask your AI assistant things like:

Translate "I would like to order a coffee, please" into French at the beginner level.

What languages does Levelang support?

Compare how "I'm worried the new rules might prevent us from finishing on time" translates into German at beginner vs advanced level.

Server Configuration

The settings below are for running the MCP server (local development or self-hosting). End users connecting via URL do not need these.

All configuration is through environment variables. When running locally via stdio, these go in the env block of your MCP client config.

LEVELANG_API_KEY is required when connecting to a remote backend (staging/production). It may be omitted for local development if the backend has auth disabled.

MCP_API_KEYS controls client authentication for the HTTP transport. When set, clients must send Authorization: Bearer <key> with a key from this list. When empty or unset, auth is disabled (open access). This has no effect on stdio transport.

Local stdio Connection

For local development you can run the MCP server as a subprocess instead of connecting via URL. This requires Python 3.12+, uv, and a running Levelang backend.

git clone https://github.com/beverage/levelang-mcp.git
cd levelang-mcp
uv sync

{
  "mcpServers": {
    "levelang": {
      "command": "uv",
      "args": [
        "run",
        "--directory", "/absolute/path/to/levelang-mcp",
        "python", "-m", "levelang_mcp"
      ],
      "env": {
        "LEVELANG_API_BASE_URL": "http://localhost:8000/api/v1"
      }
    }
  }
}

Development

Setup

uv sync
git config core.hooksPath .githooks

This enables pre-commit (auto-fix lint + format) and pre-push (lint + format check + tests) hooks.

Running Tests

uv run pytest tests/ -v

MCP Inspector

The MCP Inspector provides a web UI for browsing and invoking tools and resources:

npx @modelcontextprotocol/inspector uv run --directory /path/to/levelang-mcp python -m levelang_mcp

Project Structure

src/levelang_mcp/
├── __main__.py       # Entrypoint (python -m levelang_mcp)
├── server.py         # MCP tools and resources
├── auth.py           # API-key auth middleware for HTTP transport
├── client.py         # Async HTTP client for the Levelang API
├── config.py         # Environment variable loading
└── formatting.py     # API response → human-readable text

tests/
├── test_auth.py      # Auth middleware and config tests
├── test_client.py    # HTTP client tests (mocked)
├── test_formatting.py
└── test_tools.py     # Tool integration tests (mocked)

Architecture

MCP Client              levelang-mcp              Levelang Backend
(Claude, Cursor,   ◄── MCP/stdio ──►   (this)   ─── HTTP ──►   (FastAPI)
 Claude Code)                                     POST /translate
                                                  GET /languages/details
                                                  GET /languages/{code}

The MCP server is a stateless wrapper. It translates MCP tool calls into HTTP requests to the Levelang backend and formats responses as human-readable text for the LLM. It does not share code, database connections, or deployment with the backend.

License

MIT