txn2/mcp-trino
Full documentation at mcp-trino.txn2.com
A Model Context Protocol (MCP) server for Trino, enabling AI assistants to query and explore data warehouses with optional semantic context from metadata catalogs.
AI assistants excel at querying data but lack organizational context: which tables are trustworthy, what metrics mean, and which columns contain sensitive data. mcp-trino bridges this gap by connecting Trino to AI assistants through the MCP protocol, with an optional semantic layer that surfaces business metadata alongside query results.
Core Capabilities
Composable Architecture
- Import as a Go library to build custom MCP servers
- Add authentication, tenant isolation, audit logging without forking
- Middleware and interceptor patterns for enterprise requirements
Semantic Context
- Surface business descriptions, ownership, and data quality from metadata catalogs
- Mark sensitive columns and deprecation warnings for AI assistants
- Connect to DataHub, static files, or build custom metadata providers
Multi-Cluster Connectivity
- Query multiple Trino servers from a single MCP installation
- Unified interface across production, staging, and development environments
Secure Defaults
- Read-only mode prevents accidental data modification
- Query limits and timeouts prevent runaway operations
- SLSA Level 3 provenance for supply chain security
Features
- Execute SQL Queries: Run queries with configurable row limits and timeouts
- Analyze Execution Plans: Inspect logical, distributed, and I/O query plans
- Discover Schema: Browse catalogs, schemas, and tables across clusters
- Describe Tables: View column definitions with optional data samples
- Enrich with Context: Surface business metadata, ownership, and data quality
- Compose Custom Servers: Import as a Go library with middleware and interceptors
Installation
Homebrew (macOS)
brew install txn2/tap/mcp-trino
Claude Desktop
Claude Desktop is the GUI application for chatting with Claude. Install the mcp-trino extension to enable Trino queries in your conversations.
Option 1: One-Click Install (Recommended)
Download the .mcpb bundle for your Mac from the releases page and double-click to install:
| Mac Type | Chip | Download |
|---|---|---|
| MacBook Air/Pro (2020+), Mac Mini (2020+), iMac (2021+), Mac Studio | Apple M1, M2, M3, M4 (arm64) | mcp-trino_*_darwin_arm64.mcpb |
| MacBook Air/Pro (pre-2020), Mac Mini (pre-2020), iMac (pre-2021) | Intel (amd64) | mcp-trino_*_darwin_amd64.mcpb |
Tip: Not sure which chip you have? Click → "About This Mac". Look for "Chip" (Apple Silicon) or "Processor" (Intel).
Option 2: Manual Configuration
Add to your claude_desktop_config.json (find via Claude Desktop → Settings → Developer):
{
"mcpServers": {
"trino": {
"command": "/opt/homebrew/bin/mcp-trino",
"env": {
"TRINO_HOST": "trino.example.com",
"TRINO_USER": "your_user",
"TRINO_PASSWORD": "your_password",
"TRINO_CATALOG": "hive",
"TRINO_SCHEMA": "default"
}
}
}
}
Claude Code CLI
Claude Code is the terminal-based coding assistant. Add mcp-trino as an MCP server:
# Install via Homebrew first (see above), then:
claude mcp add trino \
-e TRINO_HOST=trino.example.com \
-e TRINO_USER=your_user \
-e TRINO_PASSWORD=your_password \
-e TRINO_CATALOG=hive \
-- mcp-trino
Or download and install manually:
# Download the latest release for your architecture
curl -L https://github.com/txn2/mcp-trino/releases/latest/download/mcp-trino_$(uname -s)_$(uname -m).tar.gz | tar xz
# Add to Claude Code
claude mcp add trino \
-e TRINO_HOST=trino.example.com \
-e TRINO_USER=your_user \
-e TRINO_PASSWORD=your_password \
-e TRINO_CATALOG=hive \
-- ./mcp-trino
Docker
docker run --rm -i \
-e TRINO_HOST=trino.example.com \
-e TRINO_USER=your_user \
-e TRINO_PASSWORD=your_password \
ghcr.io/txn2/mcp-trino:latest
Go Install
go install github.com/txn2/mcp-trino/cmd/mcp-trino@latest
Download Binary
Download pre-built binaries from the releases page. All releases are signed with Cosign and include SLSA provenance.
As a Library
go get github.com/txn2/mcp-trino
Quick Start
Multiple Trino Servers
You can configure multiple Trino instances with different names:
# Production
claude mcp add trino-prod \
-e TRINO_HOST=trino.prod.example.com \
-e TRINO_USER=prod_user \
-- mcp-trino
# Staging
claude mcp add trino-staging \
-e TRINO_HOST=trino.staging.example.com \
-e TRINO_USER=staging_user \
-- mcp-trino
Standalone Server
export TRINO_HOST=trino.example.com
export TRINO_USER=your_user
export TRINO_PASSWORD=your_password
mcp-trino
Tools
| Tool | Description |
|---|---|
trino_query | Execute SQL queries with limit/timeout control |
trino_explain | Get execution plans (logical/distributed/io/validate) |
trino_list_catalogs | List available catalogs |
trino_list_schemas | List schemas in a catalog |
trino_list_tables | List tables in a schema |
trino_describe_table | Get columns, sample data, and semantic context (if configured) |
trino_list_connections | List all configured server connections |
Semantic Layer
AI agents operate more reliably when they understand organizational context: not just table structures, but which datasets are production-ready, what business terms mean, and which columns require careful handling.
mcp-trino's semantic layer integrates with metadata catalogs to surface this context alongside query results:
| Metadata | Description |
|---|---|
| Descriptions | Business-friendly explanations of tables and columns |
| Ownership | Data stewards and technical owners |
| Tags & Domains | Classification labels and business domains |
| Glossary Terms | Links to formal business definitions |
| Data Quality | Freshness scores and quality metrics |
| Sensitivity | PII and sensitive data markers at column level |
| Deprecation | Warnings with replacement guidance |
| Lineage | Upstream and downstream data dependencies |
Providers
| Provider | Description |
|---|---|
| DataHub | Connect to DataHub's GraphQL API for enterprise metadata |
| Static Files | Load metadata from YAML or JSON files with hot-reload |
| Custom | Implement the semantic.Provider interface for any catalog |
See the Semantic Layer Documentation for configuration, caching, and custom provider development.
Configuration
| Environment Variable | Description | Default |
|---|---|---|
TRINO_HOST | Trino server hostname | localhost |
TRINO_PORT | Trino server port | 443 (SSL) / 8080 |
TRINO_USER | Authentication username | (required) |
TRINO_PASSWORD | Authentication password | (optional) |
TRINO_CATALOG | Default catalog | memory |
TRINO_SCHEMA | Default schema | default |
TRINO_SSL | Enable HTTPS | true for remote hosts |
TRINO_SSL_VERIFY | Verify SSL certificates | true |
TRINO_TIMEOUT | Query timeout (seconds) | 120 |
TRINO_SOURCE | Client identifier | mcp-trino |
TRINO_ADDITIONAL_SERVERS | Additional servers (JSON) | (optional) |
Multi-Server Configuration
Connect to multiple Trino servers from a single installation. Configure your primary server with the standard environment variables, then add additional servers via JSON:
export TRINO_HOST=prod.trino.example.com
export TRINO_USER=admin
export TRINO_PASSWORD=secret
export TRINO_ADDITIONAL_SERVERS='{
"staging": {"host": "staging.trino.example.com"},
"dev": {"host": "localhost", "port": 8080, "ssl": false}
}'
Additional servers inherit credentials and settings from the primary server unless overridden:
{
"staging": {
"host": "staging.trino.example.com",
"user": "staging_user",
"catalog": "iceberg"
},
"dev": {
"host": "localhost",
"port": 8080,
"ssl": false,
"user": "admin"
}
}
Use the connection parameter in any tool to target a specific server:
"Query the staging server: SELECT * FROM users LIMIT 10"
→ trino_query(sql="...", connection="staging")
Use trino_list_connections to discover available connections.
File-Based Configuration
For production deployments using Kubernetes ConfigMaps, Vault, or other secret management systems, mcp-trino supports file-based configuration:
# config.yaml
trino:
host: trino.example.com
port: 443
user: ${TRINO_USER} # Supports env var expansion
password: ${TRINO_PASSWORD} # Secrets can come from env
catalog: hive
schema: default
ssl: true
timeout: 120s
toolkit:
default_limit: 1000
max_limit: 10000
default_timeout: 120s
max_timeout: 300s
extensions:
logging: true
readonly: true
errors: true
Load configuration in your custom server:
import "github.com/txn2/mcp-trino/pkg/extensions"
// Load from file with env var overrides
cfg, err := extensions.LoadConfig("/etc/mcp-trino/config.yaml")
// Convert to individual configs
clientCfg := cfg.ClientConfig()
toolsCfg := cfg.ToolsConfig()
extCfg := cfg.ExtConfig()
Using as a Library
mcp-trino is designed to be composable. You can import its tools into your own MCP server:
package main
import (
"context"
"log"
"github.com/modelcontextprotocol/go-sdk/mcp"
"github.com/txn2/mcp-trino/pkg/client"
"github.com/txn2/mcp-trino/pkg/tools"
)
func main() {
// Create your MCP server
server := mcp.NewServer(&mcp.Implementation{
Name: "my-data-server",
Version: "1.0.0",
}, nil)
// Create Trino client
trinoClient, err := client.New(client.Config{
Host: "trino.example.com",
Port: 443,
User: "service_user",
SSL: true,
Catalog: "hive",
Schema: "analytics",
})
if err != nil {
log.Fatal(err)
}
defer trinoClient.Close()
// Add Trino tools to your server
toolkit := tools.NewToolkit(trinoClient, tools.Config{
DefaultLimit: 1000,
MaxLimit: 10000,
})
toolkit.RegisterAll(server)
// Add your own custom tools here...
// mcp.AddTool(server, &mcp.Tool{...}, handler)
// Run the server
if err := server.Run(context.Background(), &mcp.StdioTransport{}); err != nil {
log.Fatal(err)
}
}
Extensions
The standalone server includes optional extensions that can be enabled via environment variables:
| Environment Variable | Default | Description |
|---|---|---|
MCP_TRINO_EXT_LOGGING | false | Structured JSON logging of tool calls |
MCP_TRINO_EXT_METRICS | false | In-memory metrics collection |
MCP_TRINO_EXT_READONLY | true | Block modification statements (INSERT, UPDATE, DELETE, etc.) |
MCP_TRINO_EXT_QUERYLOG | false | Log all SQL queries for audit |
MCP_TRINO_EXT_METADATA | false | Add execution metadata footer to results |
MCP_TRINO_EXT_ERRORS | true | Add helpful hints to error messages |
Using Extensions in Custom Servers
import (
"github.com/txn2/mcp-trino/pkg/extensions"
"github.com/txn2/mcp-trino/pkg/tools"
)
// Load extension config from environment
extCfg := extensions.FromEnv()
// Or configure programmatically
extCfg := extensions.Config{
EnableLogging: true,
EnableReadOnly: true,
EnableErrorHelp: true,
}
// Build toolkit options from extensions
toolkitOpts := extensions.BuildToolkitOptions(extCfg)
// Create toolkit with extensions
toolkit := tools.NewToolkit(trinoClient, toolsCfg, toolkitOpts...)
Custom Middleware and Interceptors
You can create custom middleware, interceptors, and transformers:
// Custom middleware for authentication
authMiddleware := tools.MiddlewareFunc{
BeforeFn: func(ctx context.Context, tc *tools.ToolContext) (context.Context, error) {
// Validate user permissions
return ctx, nil
},
}
// Custom interceptor for tenant isolation
tenantInterceptor := tools.QueryInterceptorFunc(
func(ctx context.Context, sql string, toolName tools.ToolName) (string, error) {
// Add WHERE tenant_id = ? clause
return sql, nil
},
)
// Apply to toolkit
toolkit := tools.NewToolkit(client, cfg,
tools.WithMiddleware(authMiddleware),
tools.WithQueryInterceptor(tenantInterceptor),
)
Security Considerations
- Credentials: Store passwords in environment variables or secret managers
- Query Limits: Default 1000 rows, max 10000 to prevent data exfiltration
- Timeouts: Default 120s timeout prevents runaway queries
- Read-Only: ReadOnly interceptor enabled by default blocks modification statements
- Access Control: Configure Trino roles and catalog access for defense in depth
Development
# Clone the repository
git clone https://github.com/txn2/mcp-trino.git
cd mcp-trino
# Build
make build
# Run tests
make test
# Run linter
make lint
# Run all checks
make verify
# Run with a local Trino (e.g., via Docker)
make docker-trino
export TRINO_HOST=localhost
export TRINO_PORT=8080
export TRINO_USER=admin
export TRINO_SSL=false
./mcp-trino
Contributing
We welcome contributions for bug fixes, tests, and documentation. See CONTRIBUTING.md for guidelines.
License
Related Projects
- Model Context Protocol - The MCP specification
- Trino - Distributed SQL query engine
- Official Go MCP SDK - Go SDK for MCP
Open source by Craig Johnston, sponsored by Deasil Works, Inc.