Configure MCP Semantic Endpoint

The TimeXtender MCP Server endpoint implements the Model Context Protocol (MCP) to connect AI agents such as Claude, LM Studio, and other compatible tools directly to semantic models deployed in the TimeXtender Data Integration platform.

This article covers the TimeXtender Portal and TimeXtender Desktop side of the workflow. The endpoint generates a JSON representation of the semantic model that the MCP Server reads and exposes to MCP-compatible clients. AI agents use this metadata to generate SQL queries, answer natural language questions, and derive actionable insights from the data warehouse.

Prerequisites

1. Active TimeXtender Data Integration environment with Portal access
2. TimeXtender Desktop installed on the local machine
3. Prepare instance with an accessible data warehouse on one of:
   • Azure SQL Database
   • Snowflake (Early Access)
   • Microsoft Fabric (Early Access)

Create Deliver Instance with MCP Endpoint

1. Navigate to the TimeXtender Portal and access your data estate
2. Click Add endpoint
3. Select TimeXtender MCP Server endpoint from the Endpoint type dropdown
4. Enter a descriptive Name for the endpoint. This name becomes the Model Name in the MCP Configurator, so pick something meaningful for AI clients (for example, Sales, Finance, Customers).
5. Click Add endpoint to create the deliver instance
6. Specify the JSON file location where the semantic model JSON will be stored. The MCP Server reads from this path, so it must be accessible from the machine where the MCP Server is installed.
   • Enter the full file path including the .json extension
   • Example: C:\SemanticModels\Sales.json
   • The file name must use only letters, numbers, hyphens (-), and underscores (_). No spaces or special characters.
7. Configure optional View schema and Postfix settings if required
8. Save the endpoint configuration

[Screenshot: TimeXtender Portal Add endpoint dialog with TimeXtender MCP Server endpoint selected, Name field filled in, and JSON file location set.]

Build Semantic Model in TimeXtender Desktop

1. Open TimeXtender Desktop and connect to your data estate
2. Expand the Deliver instance tree and locate the MCP endpoint created above
3. Map the fields to include in the semantic model

Write Descriptions That Work for AI

Field and table descriptions directly determine AI query accuracy. The MCP Server passes these descriptions to connected AI clients, and the AI uses them to decide which tables to join, which columns to filter on, and how to interpret each measure. Clear, specific descriptions are the difference between an AI that answers consistently with your existing reports and one that drifts.

Strong descriptions answer the question "how should I use this?" rather than restating the column name. Go beyond labels to document:

• Business meaning. What this table or field represents in the business, not what type it is. Example: "A consolidated list of customers from HubSpot CRM and Business Central ERP" beats "Customer table."
• Grain. The level of detail one row represents. Example: "One row per invoice line. To get invoice totals, group by InvoiceId."
• Date logic. Which date field drives reporting for this entity, and why. Example: "Revenue is recognised on the InvoiceDate, not OrderDate. Use InvoiceDate when answering 'how much did we make in Q1?'"
• Filters and usage rules. Any filter the AI should apply by default. Example: "Exclude rows where StatusCode = 'CANCELLED' for revenue calculations."
• Measure semantics. For key measures, specify whether values represent bookings, billed amounts, or recognised revenue. Document the unit (currency, count, percentage) and any rounding rules.

To add or edit descriptions:

1. Right-click a table in the semantic model and select Edit Table
2. Enter a detailed Description that explains the business purpose and content of the table
3. Click OK to save the table description
4. Right-click individual fields and select Edit Field to add field-level descriptions
5. Repeat for all tables and critical fields in the semantic model

Tables and fields with descriptions display a caret (^) suffix in TimeXtender Desktop, indicating metadata completeness at a glance.

Treat the descriptions as living documentation. As your pilot reveals where the AI gets confused, sharpen the relevant descriptions and redeploy.

Deploy Semantic Model

1. Right-click the MCP endpoint in TimeXtender Desktop
2. Select Deploy to generate the semantic model JSON file
3. Verify the JSON file exists at the specified file location

The JSON file contains the complete schema, relationships, and descriptions. The MCP Configurator reads this file when you register the model and extracts the Model Name automatically from it.

Next Steps

• Configure MCP Server
• Manage API Keys for MCP Server