mcp

auto-ontology-mcp

An MCP server for Auto Ontology. It lets any MCP-capable agent — Cursor, Claude Desktop, an internal agent — ask questions in natural language about the data a Auto Ontology deployment is connected to, and inspect the semantic layer behind the answers.

Install and run

uvx fetches, builds, and runs it straight from the repository, so there is nothing to clone:

uvx --from "git+https://github.com/NVIDIA/auto-ontology.git#subdirectory=mcp" auto-ontology-mcp

From a checkout of this directory, uvx --from . auto-ontology-mcp does the same.

Note

Auto Ontology is NVIDIA-internal today, so this install needs GitHub credentials with access to the repository. It becomes a plain uvx auto-ontology-mcp once the package is published to PyPI.

Run one server; people log in through it. The URL of your Auto Ontology deployment is the only thing it needs to be told:

AUTO_ONTOLOGY_API_URL=https://auto_ontology.example.com auto-ontology-mcp

Connect a client

In Cursor (~/.cursor/mcp.json) or Claude Desktop (claude_desktop_config.json), point at the URL. There are no credentials in the config:

{
  "mcpServers": {
    "auto-ontology": {
      "url": "https://auto-ontology-mcp.example/mcp"
    }
  }
}

The client offers to sign in, the user gets Auto Ontology's normal login page, and every call afterwards runs as that person.

Then ask something like "what does Auto Ontology mean by an active customer, and how many were there last quarter?"

Tools

ask_question is the one that answers questions: it runs Auto Ontology's text-to-SQL agent and returns the answer, the SQL it ran, and the rows. The rest — search_terms, describe_table, check_answerable and friends — let an agent learn the vocabulary and check its assumptions first.

Every tool reads, and every tool works through the semantic layer: there is none for browsing databases, schemas, or raw columns. Nothing here modifies the glossary, the catalog, or the underlying databases.

Notes

This server is a plain HTTP client of the Auto Ontology API, so it needs no database credentials and can run anywhere that can reach your deployment. It holds no credentials of its own either — Auto Ontology signs each caller in — so it never has more access than the person calling it.

Full documentation — all configuration variables, the complete tool list, and how to extend it — is in docs/mcp.md.