Vast Db MCP

🚀 VAST Database MCP Server Implementation Explanation

This project focuses on the implementation of the MCP server for the VAST database, offering a series of resource - related and tool - based functions, and provides clear guidance on interacting with the AI agent framework.

✨ Features

1. Resources

a. List All Schemas

• Function: Display the names of all schemas in the database.
• Request Method: GET /schemas
• Response Format: A JSON array containing schema names as strings.

b. List Tables under a Specified Schema

• Function: Based on the provided schema name, list all table information under that schema.
• Request Method: GET /schemas/{schema_name}/tables
• Response Format: A JSON array where each element contains the following fields:
  • name: The name of the table.
  • type: The type of the table (e.g., BASE TABLE).
  • schema: The name of the schema the table belongs to.

c. Get Table Metadata

• Function: Based on the provided schema name and table name, return detailed metadata of the table.
• Request Method: GET /schemas/{schema_name}/tables/{table_name}
• Response Format: A JSON object containing the following fields:
  • name: The name of the table.
  • type: The type of the table.
  • schema: The name of the schema.
  • columns: An array of column information, where each column object contains:
    • name: The column name.
    • type: The data type.
    • nullable: Whether it allows null values.
    • default_value: The default value (if exists).

d. Get Table Data

• Function: Based on the provided schema name and table name, return the specific data of the table.
• Request Method: GET /schemas/{schema_name}/tables/{table_name}/data
• Response Format: A JSON array where each element corresponds to a row of data, and the fields are consistent with the column information in the metadata.

2. Tools

a. Interactive SQL Execution Tool Based on Queries

• Function: Allow users to execute arbitrary database operations by providing SQL query statements and support specifying the output format.
• Name: vast_sql_query
• Parameters:
  • sql: The SQL query string to be executed.
  • format: The output format, with optional values of json or csv.

📚 Documentation

Notes on Interacting with the AI Agent Framework

1. Prompts/Configurations

• The description of available resources and tools (similar to the "Implemented MCP Features" section) must be included in the system prompts or configurations.
• Since MCP lacks a standard discovery mechanism, all URIs, tool names, and their functions need to be explicitly listed.

2. Authentication Handling

⚠️ Important Note

Never include VAST database credentials in the prompts sent to the LLM.

• Resource Requests:

  • After the LLM generates the target URI, the orchestrator (the code running the agent logic, not the LLM) is responsible for:
    • Securely loading the stored credentials.
    • Building the headers dictionary: {'X-Vast-Access-Key': '...', 'X-Vast-Secret-Key': '...'}.
    • Mapping the vast:// URI to the actual running MCP server HTTP URL (e.g., http://localhost:8088/).
    • Sending a GET request using the MCP client or a standard HTTP client while passing the constructed headers.

• Tool Usage:

  • When the LLM decides to use the vast_sql_query tool and provides the sql and format parameters, the orchestrator is responsible for:
    • Securely loading the stored credentials.
    • Building the above headers dictionary and injecting it into the parameters of the tool execution function. The LLM should not automatically generate the headers.

3. Request Construction

• The LLM needs to generate correct URI path parameters and query parameters.
• The LLM generates the sql and format parameters for tool calls.

4. Response Handling

• The client agent code needs to handle different response MIME types (e.g., text/plain, application/json, text/csv).
• The HTTP status code must be checked (especially in resource requests) to detect errors (e.g., 401, 404, 500, 503).
• Error information (plain text or JSON) needs to be parsed from the response body and may be reported to the user or used for retry/alternative operations.

🔧 Technical Details

Next Steps

• Implement robust logging. (Completed)
• Add unit tests. (Completed)
• Fine - tune the output format. (Completed - JSON and CSV options provided)
• Enhance the error handling mechanism. (Completed)
• Improve documentation writing and examples. (In progress)

🚀 Quick Start

Start the MCP Service

uvicorn fastapi_mcp.main:app --reload

Example Requests

Get All Schemas

curl http://localhost:8000/schemas

Get Tables under a Specified Schema

curl http://localhost:8000/schemas/my_schema/tables

Get Table Metadata

curl http://localhost:8000/schemas/my_schema/tables/my_table

Get Table Data

curl http://localhost:8000/schemas/my_schema/tables/my_table/data

Example SQL Query

curl -X POST "http://localhost:8000/_api" \
     -H "Content-Type: application/json" \
     -d '{"name": "vast_sql_query", "params": {"sql": "SELECT * FROM my_table LIMIT 5", "format": "json"}}'

📄 License

Notes

• All API requests must include authentication headers.
• Appropriate permission control and logging are required for sensitive operations.