Quick start

Welcome to ScalerDeep — smart customer service and MCP plain-language analytics. Follow these steps to run your first test.

1. Configure AI models

Open **AI Settings** and enter API keys for Volcengine Ark (or another provider).

• For Volcengine chat models, use the **inference endpoint ID** (`ep-` prefix), not product names like `doubao-pro-32k`.

• After saving, pick the provider and model when editing a bot.

2. Add a data source

Open **Database Connections** and connect your business MySQL database.

• Enter host, port, credentials, and database name.

• Set an **allowed tables** whitelist (e.g. `sg_orders,sg_order_items`) so MCP can only query approved tables.

3. Prepare MCP datasets

Open **Ask Your Data (MCP)**:

• **AI generate**: pick a data source and analysis directions to batch-create semantic datasets. See [MCP AI generate](/help/mcp-ai-generate) for coverage modes.

• **Datasets**: maintain synonyms and sample questions so the system understands user phrasing.

4. Create a bot and preview

Open **AI Assistants**, create a sales or support bot, and optionally bind a knowledge base.

Click **Preview** and try sample questions:

• `Sales summary for the last 3 days`

• `Top 10 products last 3 days, show bar chart`

• `Show order data for the last three days, export excel`

5. Debug and optimize

• **Query tester**: manually call MCP tools and inspect parameters and responses.

• **Intent test** (MCP → Intent rules): type user phrases and review routing scores and dataset candidates.