Teradata MCP Server Template

Overview

The Teradata MCP server is a open source project, we welcome contributions via pull requests.

We are providing three sets of tools and associated helpful prompts

1. td_base_tools:

   • execute_read_query - runs a read query

   • execute_write_query - runs a write query

   • read_table_DDL - returns the show table results

   • read_database_list - returns a list of all databases

   • read_table_list - returns a list of tables in a database

   • read_column_description - returns description of columns in a table

   • read_table_preview - returns column information and 5 rows from the table

   • read_table_affinity - gets tables commonly used together

   • read_table_usage - Measure the usage of a table and views by users in a given schema

   • prompt_general - Create a SQL query against the database

   • prompt_table_business_description - generates a business description of a table

   • prompt_database_business_description - generates a business description of a databases based on the tables

2. td_dba_tools:

   • read_user_sql_list - returns a list of recently executed SQL for a user

   • read_table_sql_list - returns a list of recently executed SQL for a table

   • read_table_space - returns CurrentPerm table space

   • read_database_space - returns Space allocated, space used and percentage used for a database

   • read_database_version - returns the database version information

   • read_resuage_summary - Get the Teradata system usage summary metrics by weekday and hour for each workload type and query complexity bucket.

   • read_flow_control - Get the Teradata system flow control metrics by day and hour

   • read_feature_usage - Get the user feature usage metrics

   • read_user_delay - Get the Teradata user delay metrics.

   • prompt_table_archive - Create a table archive strategy for database tables.

   • prompt_database_lineage - Creates a directed lineage map of tables in a database.

3. td_data_quality_tools:

   • missing_values - returns a list of column names with missing values

   • negative_values - returns a list of column names with negative values

   • distinct_categories - returns a list of categories within a column

   • standard_deviation - returns the mean and standard deviation for a column

You may add define custom "query" tools in the custom_tools.yaml file or in any file ending with _tools.yaml. Simply specify the tool name, description and SQL query to be executed. No parameters are supported at this point.

The Test directory contains a simple ClientChatBot tool for testing tools.

Environment Set Up

Step 1 - The environment has been put together assuming you have the uv package installed on your local machine. Installation instructions for uv can be found at https://github.com/astral-sh/uv

Step 2 - Clone the mcp-server repository with

On Windows

On Mac/Linux

Step 3 - You need to update the .env file

• Rename env file to .env

• The database URI will have the following format teradata://username:password@host:1025/databasename, use a ClearScape Analytics Experience https://www.teradata.com/getting-started/demos/clearscape-analytics

  • the usename needs updating

  • the password needs updating

  • the Teradata host needs updating

  • the databasename needs updating

• LLM Credentials need to be available for /test/pydanticaiBedrock.py code to work

• SSE setting

  • SSE : Boolean to determine if your server will be using the SSE transport (SSE = True) or the stdio transport (SSE=False)

  • SSE_HOST: IP address that the server can be found at, default should be 127.0.0.1

  • SSE_PORT: Port address that the server can be fount at, default should be 8001

Example .env file

Testing your server with MCP Inspector

Step 1 - Start the server, typer the following in your terminal

NOTE: If you are running this on a Windows machine and get npx, npm or node.js errors, install the required node.js software from here: https://github.com/nodists/nodist

Step 2 - Open the MCP Inspector

• You should open the inspector tool, go to http://127.0.0.1:6274

• Click on tools

• Click on list tools

• Click on read_database_list

• Click on run

Test the other tools, each should have a successful outcome

Control+c to stop the server in the terminal

Running the server

You can simply run the server with: uv run teradata-mcp-server

Adding your sever to an Agent using stdio

Option 1 - pydanticai chatbot

     step 1 - confirm the SSE flag in .env file has been set to False

     Step 2 - Modify the ./test/ClientChatBot.py script to point to where you installed the server, you will need to modify the following line

     Step 3 - run the ./test/ClientChatBot.py script, this will create an interactive session with the agent who has access to the MCP server. From a terminal.

• Ask the agent to list the databases

• Ask the agent to list the table in a database

• Ask the agent to show all the objects in a database

• Ask the agent a question that requires SQL to run against a table

• Type "quit" to exit.

Option 2 - ADK Chatbot

     step 1 - confirm the SSE flag in .env file has been set to False

     Step 2 - move into teradata_mcp_server/test directory From a terminal.

     Step 3 - open ADK Web Server

     Step 4 - chat with the td_agent

Option 3 - mcp_chatbot

     step 0 - Modify server_config.json in the test directory, ensure path is correct.

     step 1 - confirm the SSE flag in .env file has been set to False

    Step 2 - move into teradata_mcp_server directory From a terminal and run the mcp_chatbot

    Step 3 - list the prompts by typing /prompts

    Step 4 - running a prompt to describe a database

Adding tools using stdio to Visual Studio Code Co-pilot

• confirm the SSE flag in .env file has been set to False

• In VS Code, "Show and Run Commands"

• select "MCP: Add Server"

• select "Command Stdio"

• enter "uv" at command to run

• enter name of the server for the id

• the settings.json file should open

• modify the directory path and ensure it is pointing to where you have the server installed

• add the args so that it looks like:

Note: you will need to modify the directory path in the args for your system, this needs to be a complete path. You may also need to have a complete path to uv in the command as well.

• you can start the server from within the settings.json file or you can "MCP: Start Server"

Adding tools using SSE to Visual Studio Code Co-pilot

• confirm the SSE flag in .env file has been set to False

• you need to start the server from a terminal

• In VS Code, "Show and Run Commands"

• select "MCP: Add Server"

• select "HTTP Server Sent Events"

• enter URL for the location of the server e.g. http://127.0.0.1:8001/sse

• enter name of the server for the id

• select user space

• the settings.json file should open

• add the args so that it looks like:

• within the settings.json file or you can "MCP: Start Server"

Adding the MCP server to Claude Desktop

You can add this server Claude desktop adding this entry to your claude_desktop_config.json config file:

Note: you will need to modify the directory path in the args for your system, this needs to be a complete path. You may also need to have a complete path to uv in the command as well.

Note: this requires that uv is available to Claude in your system path or installed globally on your system (eg. uv installed with brew for Mac OS users).

Exposing tools as REST endpoints with mcpo

You can use mcpo to expose this MCP tool as an OpenAPI-compatible HTTP server.

For example, using uv: uvx mcpo --port 8001 --api-key "top-secret" -- uv run teradata-mcp-server

Your Teradata tools are now available as local REST endpoints, view documentation and test it at http://localhost:8001/docs

Using the server with Open WebUI

Open WebUI is user-friendly self-hosted AI platform designed to operate entirely offline, supporting various LLM runners like Ollama. It provides a convenient way to interact with LLMs and MCP servers from an intuitive GUI. It can be integrated with this MCP server using the mcpo component.

First run mcpo as specified in the section above.

Access the UI at http://localhost:8080. To add the MCP tools, navigate to Settings > Tools > Add Connection, and enter your mcpo server connection details (eg. localhost:8001, password = top-secret if you have executed the command line in the mcpo section).

You should be able to see the tools in the Chat Control Valves section on the right and get your models to use it.

Certification