DFAgent: From Natural Language Data Interactions to Reusable Agent-Ready Tools

We present DataFoundry Agent (DFAgent), a system that forges reusable, agent-ready tools from interactive data exploration, quality, and remediation tasks. Users engage with data through natural-language prompts for operations that include inspection, transformation, and visualization. These interactions automatically generate executable code snippets that are logged. From these snippets, DFAgent acts as a foundry, synthesizing a governed catalog of enriched tools exposed via the Model Context Protocol (MCP). In this way, user-derived logic for all data operations is transformed into standardized, composable tools without reimplementation. We demonstrate how diverse interactions accumulate into a reusable toolset, highlighting a paradigm that unifies natural language interaction, executable code generation, and tool foundry processes for agentic data systems.