xCOMET MCP Server
@shuji-bonji
About xCOMET MCP Server
xCOMET MCP Server provides AI agents with the ability to evaluate machine translation quality. It integrates with the xCOMET model from Unbabel to provide:
Config
Add this server to your MCP-compatible client using the configuration below.
{
"mcpServers": {
"xcomet": {
"command": "npx",
"args": [
"-y",
"xcomet-mcp-server"
]
}
}
}Tools
No tools detected
Fetch the live tool list by running this server in a temporary sandbox using the button above.
Overview
What is xCOMET MCP Server?
xCOMET MCP Server is a translation quality evaluation server powered by the xCOMET model from Unbabel. It provides AI agents with quality scoring, error detection, and batch processing capabilities for machine translation.
How to use xCOMET MCP Server?
Install the server via npm and configure it in your MCP host (e.g., Claude Desktop) using the command npx -y xcomet-mcp-server. Use the provided tools (xcomet_evaluate, xcomet_detect_errors, xcomet_batch_evaluate) to evaluate translation quality. Optionally set environment variables for model selection, transport mode (stdio or http), and GPU acceleration.
Key features of xCOMET MCP Server
- Quality scoring on a 0–1 scale
- Error detection with minor, major, and critical severity levels
- Batch evaluation of up to 500 translation pairs
- GPU support for accelerated inference
- Persistent model loading for up to 177x speedup on consecutive requests
- Designed to integrate with other MCP servers like DeepL
Use cases of xCOMET MCP Server
- Evaluate machine translation quality for single or batch translations
- Detect and categorize translation errors with severity ratings
- Combine with translation MCP servers to create iterative quality workflows
- Quality assurance and post-editing guidance for localization projects
FAQ from xCOMET MCP Server
What models are supported and what are their memory requirements?
Three models are supported: Unbabel/XCOMET-XL (3.5B parameters, ~14GB size, ~8–10GB RAM), Unbabel/XCOMET-XXL (10.7B parameters, ~42GB size, ~20GB RAM), and Unbabel/wmt22-comet-da (580M parameters, ~2GB size, ~3GB RAM). The latter requires a reference translation.
How does the persistent server architecture work?
The server keeps the xCOMET model loaded in memory using an internal Python FastAPI server. The first request loads the model (25–90 seconds), subsequent requests run in ~500ms. Optionally, set XCOMET_PRELOAD=true to load the model at startup.
What transport modes are available?
The server supports stdio (default) and http transport modes. Set the TRANSPORT environment variable to http and optionally configure PORT (default 3000) for remote access.
How do I specify the Python executable path?
Set the XCOMET_PYTHON_PATH environment variable to an explicit Python path. Otherwise, the server auto-detects Python from pyenv, Homebrew, or the system path.
What does the quality score mean?
| Score Range | Quality | Recommendation |
|---|---|---|
| 0.9 – 1.0 | Excellent | Ready for use |
| 0.7 – 0.9 | Good | Minor review recommended |
| 0.5 – 0.7 | Fair | Post-editing needed |
| 0.0 – 0.5 | Poor | Re-translation recommended |
Frequently asked questions
What models are supported and what are their memory requirements?
Three models are supported: `Unbabel/XCOMET-XL` (3.5B parameters, ~14GB size, ~8–10GB RAM), `Unbabel/XCOMET-XXL` (10.7B parameters, ~42GB size, ~20GB RAM), and `Unbabel/wmt22-comet-da` (580M parameters, ~2GB size, ~3GB RAM). The latter requires a reference translation.
How does the persistent server architecture work?
The server keeps the xCOMET model loaded in memory using an internal Python FastAPI server. The first request loads the model (25–90 seconds), subsequent requests run in ~500ms. Optionally, set `XCOMET_PRELOAD=true` to load the model at startup.
What transport modes are available?
The server supports `stdio` (default) and `http` transport modes. Set the `TRANSPORT` environment variable to `http` and optionally configure `PORT` (default 3000) for remote access.
How do I specify the Python executable path?
Set the `XCOMET_PYTHON_PATH` environment variable to an explicit Python path. Otherwise, the server auto-detects Python from pyenv, Homebrew, or the system path.
What does the quality score mean?
| Score Range | Quality | Recommendation | |-------------|------------|----------------------------| | 0.9 – 1.0 | Excellent | Ready for use | | 0.7 – 0.9 | Good | Minor review recommended | | 0.5 – 0.7 | Fair | Post-editing needed | | 0.0 – 0.5 | Poor | Re-translation recommended |
Basic information
More Other MCP servers
Blender
ahujasidOpen-source MCP to use Blender with any LLM
Glasswarp
GlasswarpSee and control a real Windows PC you own — from any MCP client, locally or remotely. Observe (UIA + screenshots), click/type/drag/scroll, launch apps, owner Live View. BYOH: your machine, your key.
Activepieces
activepiecesAI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents
Lemon.io
lemon-ioRequest dedicated senior developers to work on your project, write job descriptions, and prep technical interviews — all without leaving your Claude chat. Just describe what you need, and Lemon.io MCP does the rest. 1–3
Inbox Zero AI MCP
elie222The world's best AI personal assistant for email. Open source app to help you reach inbox zero fast.
Comments