Installation¶

This guide will help you install Eclair and get it running on your system. Eclair supports multiple installation methods depending on your needs.

Quick Install¶

The fastest way to install Eclair is pip. Eclair works with Python 3.10 or higher.

pip install eclair

Package Availability

The Eclair package will be published to PyPI soon. In the meantime, use the development installation method below.

Development Installation¶

For the latest features or if you want to contribute to Eclair:

Clone the Repository

git clone https://github.com/mlcommons/croissant.git
cd eclair

Install Dependencies

pip install -r requirements.txt

Install in Development Mode

pip install -e .

Install Development Tools (Optional). For development with additional tools like testing and linting:

pip install -e .[dev]

You should now be able to run the Eclair server:

eclair-server

Quick Start Script¶

Eclair includes a convenient setup script:

./start.sh

This script will:

Install all required dependencies
Set up the development environment
Start the Eclair server automatically

Eclair Diagram

Configuration¶

Eclair's behavior can be customized through configuration files and environment variables.

Configuration File¶

Eclair uses config.json in the project root for its main configuration. This file controls server settings, upstream connections, and AI model preferences.

This is the default configuration.

{
  "name": "Eclair Dataset MCP Server",
  "description": "MCP server to help AI models work with datasets, built on Croissant",
  "server": {
    "host": "0.0.0.0",
    "port": 8080,
    "transport": "streamable-http"
  },
  "upstream_server": {
    "url": "https://mcp.jetty.io/mcp",
    "timeout": 5000
  },
  "gemini": {
    "model": "gemini-2.5-pro",
    "temperature": 0.3
  },
  "claude": {
    "model": "claude-3-5-sonnet-20241022",
    "temperature": 0.3,
    "max_tokens": 4096
  }
}

MCP Server settings¶

Available Transport Methods:

streamable-http - HTTP streaming (recommended)
sse - Server-sent events (deprecated)
stdio - Standard input/output (for local usage)

Upstream Server¶

Configuration for connecting to upstream MCP servers.

{
  "upstream_server": {
    "url": "https://mcp.jetty.io/mcp",  // Upstream server URL
    "timeout": 5000,                    // Connection timeout (ms)
  }
}

Compatible MCP servers

At the moment, Eclair has only been tested to work with Jetty.io, and is quite dependent on it. Over time, we hope to standardize the Eclair tools and allow multiple upstream MCP servers.

AI Agent Configuration¶

Different AI agents require specific configuration:

Gemini CLI Configuration¶

Create ~/.gemini/settings.json:

{
  "mcpServers": {
    "eclair": {
      "httpUrl": "http://localhost:8080/mcp",
      "timeout": 5000
    }
  },
  "selectedAuthType": "gemini-api-key"
}

For step by step instructions, see the Gemini CLI user guide.

Claude Code Configuration¶

Register the server:

claude mcp add --transport http Eclair http://0.0.0.0:8080/mcp

For step by step instructions, see the Claude Code user guide.

VS Code MCP Configuration¶

Add to VS Code MCP settings:

{
  "mcpServers": {
    "eclair": {
      "command": "eclair-server",
      "args": ["--transport", "stdio"]
    }
  }
}

For step by step instructions, see the IDE user guide.