Scorecard
Access Scorecard's AI model evaluation and testing tools via a Cloudflare Workers deployment.
Scorecard MCP Server on Cloudflare
This repository allows you to deploy a remote MCP server on Cloudflare Workers that enables Claude and other MCP clients to access Scorecard's evaluation tools.
Get started:
This will deploy your Scorecard MCP server to a URL like: scorecard-mcp.<your-account>.workers.dev/sse
Alternatively, you can clone this repository and deploy it using Wrangler:
git clone https://github.com/scorecard-ai/scorecard-mcp.git
cd scorecard-mcp
npm install
npm run deploy
About This MCP Server
This MCP server provides access to Scorecard's evaluation tools directly from Claude and other MCP-compatible clients. It uses Clerk for authentication and is built on Cloudflare Workers for reliable, global deployment.
The server implements the MCP specification (2025-03-26) and provides secure access to Scorecard's API for running experiments, generating synthetic data, configuring metrics, and analyzing model performance.
Connect to MCP Clients
This MCP server works with various MCP-compatible clients:
Connect to claude.ai, Cursor, and Windsurf
Once deployed, you can connect to your MCP server from Claude and other MCP-compatible clients by providing your server URL:
https://scorecard-mcp.<your-account>.workers.dev/sse
Connect via Cloudflare AI Playground
You can also connect through the Cloudflare AI Playground:
- Go to https://playground.ai.cloudflare.com/
- Enter your deployed MCP server URL (
scorecard-mcp.<your-account>.workers.dev/sse) - You can now use Scorecard's evaluation tools directly from the playground!
Connect via Claude Desktop
For local testing, you can connect to your MCP server from Claude Desktop by using the mcp-remote proxy.
Follow Anthropic's Quickstart and within Claude Desktop go to Settings > Developer > Edit Config.
Update with this configuration:
{
"mcpServers": {
"scorecard": {
"command": "npx",
"args": [
"mcp-remote",
"https://scorecard-mcp.<your-account>.workers.dev/sse" // or http://localhost:8787/sse for local testing
]
}
}
}
Restart Claude and you should see the tools become available.
Local Development
For local development, create a ".dev.vars" file with your Clerk credentials:
cp .dev.vars.example .dev.vars
Configure the following variables in your .dev.vars file:
| Variable | Source | Notes |
|---|---|---|
| CLERK_CLIENT_ID | Clerk Dashboard -> Configure -> OAuth Applications | |
| CLERK_CLIENT_SECRET | Clerk Dashboard -> Configure -> OAuth Applications | Cannot be viewed after initial generation |
| CLERK_DOMAIN | Clerk Dashboard -> Configure -> API Keys -> Frontend API URL | Override this with the Clerk development URL if using with local Scorecard server |
| CLERK_PUBLISHABLE_KEY | Clerk Dashboard -> Configure -> API Keys -> Publishable Key | Override this with the pk_test_* one if using with local Scorecard server |
Then run the development server:
npm install
npm run dev
Remember to run npx wrangler types to generate types for the environment variables.
Contributors
Special thanks to Dustin Moore for his engineering leadership in developing this MCP implementation.
Related Servers
AntBot MCP Server
A TypeScript MCP server for integrating with the AntBot AI-based RPA platform, handling tool listing and execution.
Symbolic Algebra MCP Server
Perform symbolic mathematics and computer algebra using the SymPy library.
YepCode
Execute any LLM-generated code in the YepCode secure and scalable sandbox environment and create your own MCP tools using JavaScript or Python, with full support for NPM and PyPI packages
Behavioural Prediction MCP
The Behavioural Prediction MCP Server provides AI-powered tools to analyze wallet behaviour prediction,fraud detection and rug pull prediction.
PyMOL-MCP
Enables conversational structural biology, molecular visualization, and analysis in PyMOL through natural language.
Safe Local Python Executor
A tool for safely executing local Python code without requiring external data files.
PyMilvus Code Generate Helper
Retrieves relevant code snippets and documents to assist in generating PyMilvus code, requiring a running Milvus instance.
Authless MCP Server Example
An example of a remote MCP server deployable on Cloudflare Workers without authentication.
Blueprint MCP
Browser automation via MCP for Chrome and Firefox
Remote MCP Server (Authless)
An example of a remote MCP server deployable on Cloudflare Workers without authentication.