Creativity · MCP — server

MCP Trino Server

Trino is a distributed SQL query engine that federates queries across Hive, Iceberg, Kafka, Postgres, MySQL, and dozens of connectors. The MCP Trino Server lets an LLM client list catalogs and schemas and execute federated SQL — ideal for lake-house exploration and cross-database joins.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio

Capabilities

Tools: list_catalogs, list_schemas, list_tables, run_query
Supports JWT, basic, and OAuth2 auth to Trino coordinators
Read-only role recommended

Install

pip install mcp-trino

Configuration

{
  "mcpServers": {
    "trino": {
      "command": "uvx",
      "args": ["mcp-trino"],
      "env": {
        "TRINO_HOST": "trino.example.com",
        "TRINO_PORT": "8443",
        "TRINO_USER": "claude",
        "TRINO_PASSWORD": "<password>",
        "TRINO_CATALOG": "hive"
      }
    }
  }
}

Frequently asked questions

When should I use Trino MCP vs a direct Postgres MCP?

Use Trino when you need to query across multiple warehouses or a data lake; use the Postgres MCP when your data lives in a single Postgres cluster.

Does it work with Starburst Galaxy?

Yes — Starburst is Trino-compatible. Point TRINO_HOST at your Galaxy cluster and use an access-token-based user.

Any safety recommendations?

Create a Trino user limited to specific catalogs/schemas and enforce a query timeout via session properties.

Sources

Trino Documentation — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20