Connect Stormly to MCP Clients

What You Can Connect

Use any MCP client that supports remote HTTP MCP servers and the authentication method you choose in Stormly. The current setup flow is designed for Claude.ai OAuth connections, Claude Code bearer-token setup, ChatGPT Developer Mode, Codex, Cursor, and other compatible developer tools.

• Claude.ai: add the Stormly MCP endpoint and authorize access through OAuth.
• Claude Code: use the generated command from Stormly project settings.
• ChatGPT Developer Mode: connect through remote MCP support when using compatible authentication.
• Codex, Cursor, and other MCP clients: use the endpoint, token, or OAuth flow supported by that client.

Step 1 - Open MCP Settings

1. Open the Stormly application.
2. Select the project you want the AI client to query.
3. Open project settings and choose MCP Settings.

MCP credentials are project scoped. A client connected to one project cannot query another project through the same endpoint.

Step 2 - Choose OAuth or Setup Token

Stormly supports two setup styles. Use OAuth when the client can authorize a remote MCP server directly. Use a setup token when the client expects a URL plus bearer-token header.

Authorization: Bearer YOUR_STORMLY_MCP_TOKEN

The Stormly UI shows setup tokens, connected OAuth clients, and the project-specific endpoint. Admins can revoke setup tokens or OAuth client connections from the same screen.

Step 3 - Add the MCP Endpoint

The endpoint uses your Stormly project UUID. The exact URL is generated inside project settings, but it follows this structure:

https://data.stormly.com/mcp/YOUR_PROJECT_UUID

For Claude Code, copy the generated command from Stormly:

claude mcp add --transport http stormly https://data.stormly.com/mcp/YOUR_PROJECT_UUID \
  --header "Authorization: Bearer YOUR_STORMLY_MCP_TOKEN"

Step 4 - Ask One Complete Analytics Question

Ask one complete ecommerce analytics objective at a time. Include date ranges, filters, breakdowns, and dimensions in the same question when they belong together.

"Compare revenue, units sold, and cart abandonment by product category for paid search visitors last week versus the previous week."

Stormly accepts the question, returns an analysis ID, runs the analysis, and then the MCP client polls for the final answer.

What Stormly Exposes Through MCP

• submit_question submits the raw ecommerce analytics question and returns an analysis ID.
• get_answer_result polls that analysis ID until Stormly returns the final answer.
• The server instructions tell clients to send the original analytics request without paraphrasing.
• Stormly automatically chooses between a matching ecommerce report and custom SQL.

Good Questions for Ecommerce MCP

• Which SKUs created the most revenue from repeat buyers in the last 30 days?
• Which products were added to cart often but purchased rarely?
• Which campaigns drove higher AOV but lower checkout completion?
• Which categories have the largest recoverable abandoned-cart revenue?
• Which new arrivals outperformed during their first week live?