Use OuterProduct from Claude, Cursor, and MCP Clients

The OuterProduct MCP server bridges your favourite LLM interface and the OuterProduct platform. Once you add a single URL, the assistant can train models on structured data, generate predictions, produce per-feature reasoning, run counterfactuals, and mine decision-pattern aggregates, all without leaving your chat window.

What the server exposes

When connected, the assistant gains access to five categories of tools:

Training

Fit a model on any structured dataset you upload or reference.

Prediction

Score new rows against a trained model.

Explanation

Get per-feature importance for any individual prediction.

Counterfactual

Find the minimal realistic changes that flip an outcome.

Upload

Send CSV data straight from the conversation.

Pattern aggregates

Mine decision-pattern summaries across a dataset.

Server details

Property	Value
Server URL	`https://mcp.outerproduct.com/mcp`
Authentication	OAuth 2.1, no API key required
Hosting	Remote (no local process to manage)

Authentication

The server uses OAuth 2.1, so there is no API key to copy and paste. Modern MCP clients (Claude Desktop, Claude Code, and Cursor) drive the authentication flow automatically. You add the server URL, click Authenticate, sign in to OuterProduct in your browser, and you’re done.

Connect your client

Choose the tab for the client you are using.

Claude Desktop
Claude Code
Cursor

Open Settings

In Claude Desktop, go to Settings → Connectors → Add custom connector.

Paste the server URL

Enter the following URL and confirm:

https://mcp.outerproduct.com/mcp

Authenticate

Claude Desktop opens a browser tab. Sign in to OuterProduct and grant access. You are redirected back to Claude Desktop automatically.

Run this command in your terminal:

claude mcp add --transport http outerproduct https://mcp.outerproduct.com/mcp

Authenticate

Inside Claude Code, run /mcp, select outerproduct from the list, then choose Authenticate. Claude Code opens a browser tab to complete the sign-in. The same flow triggers automatically the first time the assistant calls an OuterProduct tool.

Edit mcp.json

Open ~/.cursor/mcp.json in any text editor and add the outerproduct entry:

{
  "mcpServers": {
    "outerproduct": {
      "url": "https://mcp.outerproduct.com/mcp"
    }
  }
}

If the file does not exist yet, create it with the content above.

Restart Cursor

Quit and relaunch Cursor so it picks up the new configuration.

Authenticate

Navigate to Settings → MCP → outerproduct and click Authenticate. Cursor opens a browser tab where you sign in to OuterProduct.

Verify the connection

Once authentication completes, ask the assistant:

What MCP tools do you have available from outerproduct?

A healthy connection returns a list that includes training, prediction, explanation, counterfactual, and upload tools. Anything you can do in Python with the OuterProduct SDK, you can now ask the assistant to do in plain English.

If the assistant does not list OuterProduct tools, confirm that you completed the Authenticate step for your client and that you are signed in to the correct OuterProduct account.

Built-in workflows

OuterProduct ships pre-built, end-to-end workflows for common scenarios — such as Suspicious Activity Reports (SAR) and Churn Winback Briefings — that the assistant follows automatically when your prompt matches. See Sample Prompts for the full list and the prompts that trigger them.

​What the server exposes

Training

Prediction

Explanation

Counterfactual

Upload

Pattern aggregates

​Server details

​Authentication

​Connect your client

​Verify the connection

​Built-in workflows

What the server exposes

Server details

Authentication

Connect your client

Verify the connection

Built-in workflows