Scorecard

You can run Scorecard’s MCP server on Cloudflare Workers to expose Scorecard’s evaluation tools to Claude and other MCP clients. It enables secure, global access to experiments, synthetic data generation, metric configuration, and model performance analysis through MCP-compatible clients.

How to use

After you deploy your MCP server, you can connect MCP clients to it to run evaluations, configure metrics, and analyze model performance. Use the HTTP endpoint for remote clients or the local stdio configuration to run the server from your development environment. Start by choosing the connection method that fits your workflow, then follow the setup steps to authenticate and begin sending requests to Scorecard’s evaluation tools.

How to install

Prerequisites: you need Node.js and npm installed on your machine to deploy and run the MCP server locally. You will also use Wrangler for Cloudflare Workers deployment.

Step 1: Deploy to Cloudflare Workers via the remote deployment flow.

# Deploy using the remote flow (URL provided to initiate deployment)
# You will land on a URL that deploys the MCP server to your Cloudflare Workers account

Step 2: Or clone the project locally and deploy with Wrangler.

git clone https://github.com/scorecard-ai/scorecard-mcp.git
cd scorecard-mcp
npm install
npm run deploy

Additional setup notes

For local development and testing, you can run a development server and use Clerk credentials. Create a local configuration file named ".dev.vars" based on the example, then start the dev server.

cp .dev.vars.example .dev.vars

Available tools

authenticate

Authenticate users via Clerk to control access to MCP endpoints.

deploy

Deploy the MCP server to Cloudflare Workers using the remote flow or Wrangler locally.

dev

Run a local development server to test MCP interactions and tooling.

types

Generate TypeScript types for environment variables using the Wrangler types command.