Skip to content

mcp.md

Latest commit

 

History

History
76 lines (56 loc) · 1.91 KB

mcp.md

File metadata and controls

76 lines (56 loc) · 1.91 KB

MCP Server Integration

IceFrame includes a Model Context Protocol (MCP) server that exposes its capabilities to AI assistants and IDEs (like Claude Desktop, Cursor, etc.) via stdio.

Installation

Install IceFrame with the mcp extra:

pip install "iceframe[mcp]"

Configuration

The MCP server requires the same environment variables as the IceFrame library to connect to your Iceberg catalog.

  • ICEBERG_CATALOG_URI (Required)
  • ICEBERG_CATALOG_TYPE (Default: rest)
  • ICEBERG_WAREHOUSE
  • ICEBERG_TOKEN
  • ICEBERG_CREDENTIAL
  • ICEBERG_OAUTH2_SERVER_URI

Usage

Getting Configuration

To get the JSON configuration for your MCP client (e.g., for claude_desktop_config.json), run:

iceframe mcp config

This will output a JSON object like:

{
  "mcpServers": {
    "iceframe": {
      "command": "/path/to/python",
      "args": [
        "-m",
        "iceframe.cli",
        "mcp",
        "start"
      ],
      "env": {
        "ICEBERG_CATALOG_URI": "..."
      }
    }
  }
}

Copy this configuration into your client's settings file.

Starting the Server

The server is typically started automatically by the MCP client using the command specified in the configuration. However, you can start it manually for testing:

iceframe mcp start

Available Tools

The MCP server exposes the following tools to the AI assistant:

  • list_tables(namespace): List tables in a namespace.
  • describe_table(table_name): Get schema and metadata for a table.
  • get_table_stats(table_name): Get table statistics.
  • execute_query(table_name, query, limit): Execute a query (filter expression) on a table.
  • generate_code(operation): Generate Python code for complex operations.
  • generate_sql(description): Generate SQL query templates.
  • list_documentation(): List available documentation files.
  • read_documentation(page): Read the content of a documentation file.