Overview
After deploying your Semantic Model MCP Endpoint, install the TimeXtender MCP Server and use the TimeXtender MCP Configurator to register your semantic models, generate API keys for AI clients, and start the server.
The MCP Server runs as a single Windows service that hosts every semantic model you register. AI clients reach all of them through one endpoint URL and authenticate with an API key scoped to the models that client should see.
Prerequisites
- Configure MCP Semantic Endpoint
- Read-only database user credentials for each data warehouse you want to expose
Install TimeXtender MCP Server
- Complete the Early Access Sign-up
- Download the TimeXtender MCP Installer
- Run the installer executable
- Follow the installation wizard prompts
- Wait for the TimeXtender MCP Configurator to launch automatically when installation completes
Configure the Prepare Instance Connection
Each semantic model is backed by a data warehouse. Add at least one provider before registering models. You can configure more than one provider type on the same MCP Server (for example, Azure SQL for one model and Snowflake for another).
- Open the Prepare Instance Config tab in the TimeXtender MCP Configurator
- Select the Database Type for your data warehouse:
- Azure SQL Database
- Snowflake (Early Access)
- Microsoft Fabric (Early Access)
- Enter the connection details for the provider you chose. See the matching section below.
- Click Test Connection to verify connectivity
- Click Save once the connection test succeeds
Use a read-only database user account for each provider. While the MCP server does not support ALTER, UPDATE, or DELETE statements, AI behavior can be unpredictable. Read-only credentials ensure data warehouse reliability.
Azure SQL Database
- Select the Authentication Mode:
- SQL Authentication for a SQL Server login
- Entra ID (Service Principal) for an Azure AD application
- Enter the Server address (example:
myserver.database.windows.net) - Enter the Database name (example:
mydb) - For SQL Authentication, enter the User ID and Password
- For Entra ID (Service Principal), enter the Tenant ID, Client ID, and Client Secret
The password and client secret are encrypted with Windows DPAPI before being stored on disk.
Snowflake (Early Access)
Snowflake supports three authentication modes: key pair (unencrypted .p8), key pair (encrypted .p8), and password (legacy). Key-pair authentication is recommended for production.
For the full Snowflake setup, including how to generate the key pair, associate the public key with the Snowflake user, and configure file permissions for the private key, see Connect Prepare Instance to Snowflake.
Microsoft Fabric (Early Access)
Microsoft Fabric requires Entra ID (Service Principal) authentication. SQL Authentication is not available for Fabric.
For the full Fabric setup, including how to register the service principal in Microsoft Entra, grant it access to the Fabric workspace, and find the SQL analytics endpoint hostname, see Connect Prepare Instance to Microsoft Fabric.
Add a Semantic Model
The Models tab is where you register each semantic model JSON file produced by a Deliver instance MCP endpoint. Register one model per semantic layer file. You can register as many models as you need on a single MCP Server.
- Click the Models tab
- Click Add Model
- In the Semantic Layer File field, browse to or paste the full path to the semantic model JSON file (example:
C:\TimeXtender\sales.json). The Model Name is extracted automatically from the file. - In the Provider dropdown, select the data warehouse provider this model queries. Only providers that have been saved and successfully tested in the Prepare Instance Config tab appear here.
- Click Test Connection to confirm the model can reach its data
- Click Save All to persist the model
Models can target different providers. For example, a Sales model on Azure SQL and a Production model on Snowflake can run on the same MCP Server simultaneously.
Generate API Keys for HTTP Clients
AI clients that connect over HTTP mode authenticate with an API key. Each key is a bundle that can grant access to one model, several models, or every model on the server. Keys are stored as SHA-256 hashes in the configuration file and never re-displayed in plaintext after generation.
For the full workflow, see Manage API Keys for MCP Server.
AI clients using Stdio mode (such as Claude Desktop running on the same machine) do not need API keys.
(Optional) Register with TimeXtender Cloud
If you want Xpilot or other cloud-hosted AI tools to call this MCP Server, register the server with the TimeXtender Data Platform. See Register MCP Server with TimeXtender Cloud.
Start the MCP Server
- Click the Service Management tab
- In the Endpoint section, enter the Canonical URI that AI clients will use to reach this server (example:
https://mcp.contoso.comorhttps://localhost:6001). The Canonical URI is also used for HTTPS certificate hostname validation. - (Recommended for production) Configure HTTPS in the HTTPS Certificate section. See Configure HTTPS for MCP Server.
- Click Save & Apply to persist the configuration
- Click Start to start the Windows service
- Copy the MCP Client URL from the Endpoint section. This is the URL your AI clients paste into their configuration.
Troubleshooting
Provider does not appear in the Models tab dropdown
Only providers that have been saved and successfully tested in Prepare Instance Config are listed. Return to that tab, run Test Connection, and click Save.
Test Connection fails for Azure SQL
Verify firewall rules allow connections to the database server from this machine. Confirm the server address, database name, and credentials are correct. For Entra ID (Service Principal), confirm the service principal has been granted read access to the database.
Test Connection fails for Snowflake or Microsoft Fabric
See the provider-specific troubleshooting in Connect Prepare Instance to Snowflake or Connect Prepare Instance to Microsoft Fabric for the full list of failure modes and how to diagnose them.
Invalid JSON file path error in the Models tab
The semantic model JSON file name must use only letters, numbers, hyphens (-), and underscores (_). Remove any spaces and special characters. Confirm the file uses the .json extension.
Service fails to start
Open Service Management and click Refresh to confirm the current state. Common causes: the Canonical URI port is already in use by another process, the HTTPS certificate has expired, or the service account does not have permission to read the semantic model JSON files. Click Open Logs Folder for the service log.
MCP Client URL is empty or shows the wrong host
The MCP Client URL is derived from the Canonical URI. Enter the public URL clients will use, then click Save & Apply.
