CData Connect AI — Base Skill

This skill governs how to use the CData Connect AI MCP server to answer business questions against live enterprise data. It encodes the discovery workflow the MCP requires, plus the error-recovery patterns that resolve the most common failure modes.

When NOT to use

• The user uploaded a file (CSV, Excel, JSON) and wants analysis of that file — use a separate data analysis skill instead
• The user wants to configure connections, manage OAuth, or administer Connect AI — direct them to the CData admin UI
• The user is asking about Connect AI's architecture, pricing, or strategy — this is a product conversation, not a query task

Step 0 — Confirm tool availability

Call tool_search to verify Connect AI tools exist. Search for "getCatalogs" or "queryData" for generic MCP servers, or "get_tables" or "query_data" for toolkit configurations. Do not search for the data source name (e.g., "Jira", "Salesforce") — Connect AI tools are named by function, not by data source. MCP server references in system context or artifact configurations do NOT count as confirmation. If tool_search finds nothing, stop — the server is not connected.

If the server is not connected:

• Do not attempt workarounds such as building artifacts, fetching the MCP endpoint directly, or using web_fetch against the Connect AI API
• Do inform the user that the CData Connect AI MCP server is not currently connected
• Do suggest they enable it in their MCP settings, connected apps, or integration settings
• Do not infer that Connect AI is connected based on MCP server references in system context or artifact configurations. Only confirm availability through tool_search results.