✨ MCP server for AI agents

Overview

The IOMETE MCP Server exposes the lakehouse to AI agents through the Model Context Protocol (MCP) — the emerging open standard for connecting AI assistants and agents to external tools and data. Instead of building a bespoke integration for each AI client, teams point any MCP-compatible agent (Claude, and other MCP hosts) at IOMETE and let it interact with lakehouse data through a standard interface. The capability rolls out in phases: the first iteration lets agents run and execute SQL statements against IOMETE, and a following iteration moves the Data Catalog behind MCP so agents can discover and understand available data, not just query it.

Features

Phase 1 — SQL execution over MCP

• SQL execution tool — AI agents can run SQL statements against IOMETE compute through the MCP server and receive results back in a structured form.

Phase 2 — Data Catalog behind MCP

• Catalog & metadata as MCP resources — Databases, tables, schemas, and descriptions are exposed through MCP so agents can browse and understand what data exists.
• Schema-aware querying — Agents discover the relevant tables and columns before generating SQL, improving accuracy and reducing hallucinated queries.

Across both phases

• Governed access — Designed to operate within IOMETE's existing authorization model (RAS), so agent access is subject to the same permissions and policies as other consumers. (Confirm exact enforcement behavior before publishing.)

Use cases

• Conversational analytics — Users ask questions in natural language and the agent runs the underlying SQL against the lakehouse, returning answers in context.
• Agentic data workflows — AI agents autonomously explore data, compose queries, and chain steps as part of larger tasks.
• Grounded NL-to-SQL — With the catalog behind MCP, agents ground their queries in the real schema instead of guessing table and column names.
• Embed IOMETE into AI apps — Product teams plug the lakehouse into AI assistants and copilots through a single standard protocol.

‍