The mcp-server-trino is an MCP server designed to connect AI models with data stored in Trino, a distributed SQL query engine. It exposes Trino tables as MCP resources, enabling AI models to seamlessly access and query data. This server provides a tool to execute arbitrary SQL queries against Trino, allowing for flexible data retrieval and manipulation. By leveraging Trino's Python client, the server facilitates a connection to a Trino host, catalog, and schema, making big data analytics accessible to AI models. The core value lies in simplifying the integration of AI models with large datasets managed by Trino, streamlining data access and analysis workflows. It is implemented in Python, utilizing the trino and mcp libraries, and configured via environment variables for Trino connection details.

Expose Trino Tables as Resources

The mcp-server-trino exposes Trino tables as MCP resources, allowing AI models and other components within the MCP ecosystem to discover and interact with these tables in a standardized way. This functionality bridges the gap between Trino's data analytics capabilities and the broader AI landscape. The server queries the Trino metadata to identify available tables within the configured catalog and schema. Each table is then represented as an MCP resource, complete with metadata such as column names and data types. This allows AI models to understand the structure of the data before querying it, facilitating tasks like automated feature engineering or data validation. For example, an AI model designed to predict customer churn could use this feature to automatically discover and access customer data stored in a Trino table, without needing hardcoded table names or schema definitions. The server leverages Trino's Python client (trino.dbapi) to retrieve table metadata.

Read Table Contents via MCP

This feature enables reading the contents of Trino tables through the MCP protocol. Once a table is exposed as an MCP resource, authorized clients can request the data within that table. The server then executes a query against Trino to retrieve the data and streams it back to the client in a format compatible with the MCP standard. This allows AI models to directly access and process data stored in Trino without needing to implement custom data connectors or deal with Trino's specific API. For instance, a natural language processing model could use this feature to access a large dataset of text documents stored in Trino, enabling it to train on a massive corpus of text data. The server uses standard SQL queries via the trino.dbapi to fetch data, and handles pagination and data formatting to ensure efficient data transfer.

Execute Arbitrary SQL Queries

The mcp-server-trino provides a tool for executing arbitrary SQL queries against Trino. This allows clients to leverage the full power of Trino's SQL engine through the MCP protocol. This feature is particularly useful for complex data transformations, aggregations, or filtering operations that are difficult or impossible to perform directly within the AI model. The client sends an SQL query to the server, which then executes it against Trino and returns the results. This allows AI models to delegate data processing tasks to Trino, improving performance and reducing the computational burden on the model itself. For example, a machine learning model could use this feature to pre-aggregate data in Trino before training, reducing the size of the training dataset and improving training speed. The server uses the trino.dbapi to execute the provided SQL query and returns the result set to the client. Security considerations are paramount; access control and query validation should be implemented to prevent unauthorized access or malicious queries.