loading…
Search for a command to run...
loading…
A Go implementation of a Model Context Protocol (MCP) server for Trino
A Go implementation of a Model Context Protocol (MCP) server for Trino
A high-performance Model Context Protocol (MCP) server for Trino implemented in Go. This project enables AI assistants to seamlessly interact with Trino's distributed SQL query engine through standardized MCP tools.
GitHub Workflow Status Go Version Trivy Scan SLSA 3 Go Report Card Go Reference Docker Image GitHub Release License: MIT
This project implements a Model Context Protocol (MCP) server for Trino in Go. It enables AI assistants to access Trino's distributed SQL query engine through standardized MCP tools.
Trino (formerly PrestoSQL) is a powerful distributed SQL query engine designed for fast analytics on large datasets.
graph TB
subgraph "AI Clients"
CC[Claude Code]
CD[Claude Desktop]
CR[Cursor]
WS[Windsurf]
CW[ChatWise]
end
subgraph "Authentication (Optional)"
OP[OAuth Provider<br/>Okta/Google/Azure AD]
JWT[JWT Tokens]
end
subgraph "MCP Server (mcp-trino)"
HTTP[HTTP Transport<br/>/mcp endpoint]
STDIO[STDIO Transport]
AUTH[OAuth Middleware]
TOOLS[MCP Tools<br/>• execute_query<br/>• list_catalogs<br/>• list_schemas<br/>• list_tables<br/>• get_table_schema<br/>• explain_query]
end
subgraph "Data Layer"
TRINO[Trino Cluster<br/>Distributed SQL Engine]
CATALOGS[Data Sources<br/>• PostgreSQL<br/>• MySQL<br/>• S3/Hive<br/>• BigQuery<br/>• MongoDB]
end
%% Connections
CC -.->|OAuth Flow| OP
OP -.->|JWT Token| JWT
CC -->|HTTP + JWT| HTTP
CD -->|STDIO| STDIO
CR -->|HTTP + JWT| HTTP
WS -->|STDIO| STDIO
CW -->|HTTP + JWT| HTTP
HTTP --> AUTH
AUTH -->|Validated| TOOLS
STDIO --> TOOLS
TOOLS -->|SQL Queries| TRINO
TRINO --> CATALOGS
%% Styling
classDef client fill:#e1f5fe
classDef auth fill:#f3e5f5
classDef server fill:#e8f5e8
classDef data fill:#fff3e0
class CC,CD,CR,WS,CW client
class OP,JWT auth
class HTTP,STDIO,AUTH,TOOLS server
class TRINO,CATALOGS data
Key Components:
X-Trino-Client-Tags/Info headersX-Trino-User headerInstall:
# Homebrew
brew install tuannvm/mcp/mcp-trino
# Or one-liner (macOS/Linux)
curl -fsSL https://raw.githubusercontent.com/tuannvm/mcp-trino/main/install.sh | bash
Run (Local Development):
export TRINO_HOST=localhost TRINO_USER=trino
mcp-trino
For production deployment with OAuth, see Deployment Guide and OAuth Architecture.
mcp-trino can be used as an interactive CLI similar to psql or the Trino CLI:
# Interactive REPL mode
mcp-trino --interactive
# Execute a query directly
mcp-trino query "SELECT * FROM my_table LIMIT 10"
# List catalogs, schemas, tables
mcp-trino catalogs
mcp-trino schemas my_catalog
mcp-trino tables my_catalog my_schema
# Describe a table
mcp-trino describe my_catalog.my_schema.my_table
# Explain a query
mcp-trino explain "SELECT COUNT(*) FROM my_table"
# Output formats
mcp-trino --format json query "SELECT 1"
mcp-trino --format csv query "SELECT 1"
mcp-trino --format table query "SELECT 1" # default
Every command has structured, LLM-friendly help output:
# Main help with all commands, flags, examples, and environment variables
mcp-trino --help
# Per-subcommand help
mcp-trino query --help
mcp-trino describe --help
Help output follows Unix man-page conventions with sections: NAME, SYNOPSIS, DESCRIPTION, COMMANDS, FLAGS, EXAMPLES, ENVIRONMENT, and CONFIGURATION.
| Code | Meaning |
|---|---|
| 0 | Success |
| 1 | Runtime error (connection failed, query error, etc.) |
| 2 | Usage error (unknown command, invalid flags, missing arguments) |
mcp-trino supports named connection profiles for easy switching between Trino environments.
Configuration File — supports both YAML (~/.config/trino/config.yaml) and JSON (~/.config/trino/config.json):
# ~/.config/trino/config.yaml
current: prod
profiles:
prod:
host: trino.example.com
port: 443
user: prod_user
password: prod_password
catalog: hive
schema: analytics
ssl:
enabled: true
insecure: false
dev:
host: localhost
port: 8080
user: trino
catalog: memory
schema: default
staging:
host: staging-trino.example.com
port: 443
user: staging_user
output:
format: table
Or equivalently in JSON:
{
"current": "prod",
"profiles": {
"prod": {
"host": "trino.example.com",
"port": 443,
"user": "prod_user",
"catalog": "hive",
"ssl": { "enabled": true }
},
"dev": {
"host": "localhost",
"port": 8080,
"user": "trino"
}
},
"output": { "format": "table" }
}
When both files exist, config.json takes precedence. New configs default to JSON.
Profile Management Commands:
# List all profiles
mcp-trino config profile list
# Set default profile
mcp-trino config profile use prod
# Show profile details
mcp-trino config profile show staging
# Use a specific profile (overrides config file)
mcp-trino --profile dev catalogs
Configuration Precedence (highest to lowest):
--host, --port, etc.)--profile flagTRINO_PROFILE environment variablecurrent field in config filedefault profile fallbackTRINO_HOST, etc.)Environment Variables (lowest priority - overridden by profiles and flags):
export TRINO_HOST=trino.example.com
export TRINO_PORT=443
export TRINO_USER=myuser
export TRINO_PASSWORD=mypass
export TRINO_CATALOG=hive
export TRINO_SCHEMA=analytics
export TRINO_SSL=true
Secret Management (recommended):
Secrets are loaded purely from environment variables. Use a secrets CLI to inject them via Unix piping at launch time — the app never touches your vault:
# 1Password CLI — resolves op:// references in an env file
op run --env-file=.env -- mcp-trino
# Or inline per-variable
TRINO_PASSWORD=$(op read 'op://Engineering/Trino/password') mcp-trino
See docs/secrets.md for 1Password, Vault, and Kubernetes patterns, and for security nuances (shell-history, process-list, and env-var leakage).
REPL Meta-Commands (in interactive mode):
\help - Show help\quit, \exit, \q - Exit REPL\history - Show command history\catalogs - List all catalogs\schemas [catalog] - List schemas\tables [catalog schema] - List tables\describe <table> - Describe table\format <table|json|csv> - Change output formatSupported Clients: Claude Desktop, Claude Code, Cursor, Windsurf, ChatWise
Available Tools: execute_query, list_catalogs, list_schemas, list_tables, get_table_schema, explain_query
For client integration and tool documentation, see Integration Guide and Tools Reference.
Key Variables: TRINO_HOST, TRINO_USER, TRINO_SCHEME, MCP_TRANSPORT, OAUTH_PROVIDER
Secret Management: Inject secrets through the process environment — mcp-trino reads them directly. See docs/secrets.md for 1Password, Vault, and Kubernetes recipes.
# 1Password (biometric-gated, zero disk writes)
op run --env-file=.env -- mcp-trino
# Vault (via vault-agent or CLI)
TRINO_PASSWORD=$(vault kv get -field=password secret/mcp-trino) mcp-trino
# Kubernetes: use standard Secret → envFrom in the Helm chart values
OAuth Configuration:
# Native mode (most secure - zero server-side secrets)
export OAUTH_ENABLED=true OAUTH_MODE=native OAUTH_PROVIDER=okta
export OIDC_ISSUER=https://company.okta.com OIDC_AUDIENCE=https://mcp-server.com
# Proxy mode (centralized credential management)
export OAUTH_MODE=proxy OIDC_CLIENT_ID=app-id OIDC_CLIENT_SECRET=secret
export OAUTH_REDIRECT_URI=https://mcp-server.com/oauth/callback # Fixed mode (localhost-only)
export OAUTH_REDIRECT_URI=https://app1.com/cb,https://app2.com/cb # Allowlist mode
export JWT_SECRET=$(openssl rand -hex 32) # Required for multi-pod deployments
Performance Optimization:
# Focus AI on specific schemas only (10-20x performance improvement)
export TRINO_ALLOWED_SCHEMAS="hive.analytics,hive.marts,hive.reporting"
User Identity Tracking:
# Query Attribution is AUTOMATIC when OAuth is enabled
# Queries are tagged with X-Trino-Client-Tags and X-Trino-Client-Info headers
# For full impersonation (Trino enforces user permissions):
export TRINO_ENABLE_IMPERSONATION=true
export TRINO_IMPERSONATION_FIELD=email # Options: username, email, subject
For complete configuration, see Deployment Guide, OAuth Guide, Allowlists Guide, and User Identity Guide.
mcp-trino uses oauth-mcp-proxy - a standalone OAuth 2.1 library for Go MCP servers.
Why a separate library?
For OAuth details:
Contributions are welcome! Please feel free to submit a Pull Request.
This project is licensed under the MIT License - see the LICENSE file for details.
This project uses GitHub Actions for continuous integration and GoReleaser for automated releases.
Our CI pipeline performs the following checks on all PRs and commits to the main branch:
When changes are merged to the main branch:
Добавь это в claude_desktop_config.json и перезапусти Claude Desktop.
{
"mcpServers": {
"tuannvm-mcp-trino": {
"command": "npx",
"args": []
}
}
}