MCP Server¶
The mcp-trino server enables AI assistants to query Trino data warehouses through the Model Context Protocol (MCP).
What It Does¶
When connected to your AI assistant, the server provides tools for:
- SQL Execution - Run queries with automatic limits and timeouts
- Query Analysis - Explain execution plans before running expensive queries
- Schema Exploration - Browse catalogs, schemas, tables, and columns
- Multi-Cluster - Query multiple Trino clusters from one installation
Supported AI Clients¶
| Client | Platform | Install Method |
|---|---|---|
| Claude Desktop | macOS, Windows | .mcpb bundle |
| Claude Code | CLI | claude mcp add |
| Cursor | macOS, Windows, Linux | Manual config |
| Windsurf | macOS, Windows, Linux | Manual config |
| Any MCP Client | Any | stdio transport |
Quick Demo¶
Once installed, ask your AI assistant:
"What tables are in the sales schema?"
The assistant uses trino_list_tables to show available tables.
"Show me the top 10 customers by order value"
The assistant generates SQL and uses trino_query to execute it.
"Why is my orders query slow?"
The assistant uses trino_explain to analyze the execution plan.
Features¶
Secure Defaults¶
- Read-only mode enabled by default (blocks INSERT, UPDATE, DELETE)
- Row limits prevent excessive data retrieval (default: 1000, max: 10000)
- Query timeouts prevent runaway queries (default: 120s)
Production Ready¶
- File-based configuration for Kubernetes ConfigMaps
- Environment variable expansion for secrets
- Multi-server support for environment separation
Enterprise Security¶
- SLSA Level 3 provenance on all releases
- Cosign keyless signatures for verification
- Checksum verification
Next Steps¶
- Installation - Install the server
- Configuration - Configure connection and extensions
- Tools - Learn about available tools
- Multi-Server - Connect multiple clusters