agent/

LLM Integration for TuiML.

Provides MCP (Model Context Protocol) server and tools for LLM integration. Workflow/discovery tools are exposed as MCP tools, and ALL 200+ components are accessible through them. New algorithms registered via decorators or the component registry are automatically discoverable.

Quick Start

Run the MCP server:

python -m tuiml.agent.mcp.server

Or configure in Claude Desktop (claude_desktop_config.json):

{
    "mcpServers": {
        "tuiml": {
            "command": "tuiml-mcp"
        }
    }
}

Exposed MCP Tools

Workflow Tools:

tuiml_train - Train ML models with full workflow
tuiml_predict - Make predictions
tuiml_evaluate - Evaluate models
tuiml_benchmark - Compare algorithms
tuiml_upload_data - Upload dataset content

Discovery Tools:

tuiml_list - List all components; pass search= to filter by keyword
tuiml_describe - Get component details and parameter schema

All 200+ algorithms, preprocessors, datasets, features, and splitters are accessible through these tools. When a new algorithm is added to the component registry or registered with a decorator, it is automatically available.

python

>>> from tuiml.agent import execute_tool
>>>
>>> # Train any algorithm by name
>>> result = execute_tool(
...     "tuiml_train",
...     algorithm="RandomForestClassifier",
...     data="iris",
...     target="class",
...     cv=10
... )
>>>
>>> # List available algorithms
>>> result = execute_tool("tuiml_list", category="algorithm")

Packages