概要
What is Scorable MCP?
Scorable MCP is a Model Context Protocol (MCP) server that bridges the Scorable API with MCP client applications. It exposes Scorable evaluators as tools for AI assistants and agents, enabling automated evaluation of responses against various quality criteria.
How to use Scorable MCP?
Get an API key from Scorable, then run the server using Docker with SSE transport (docker run -e SCORABLE_API_KEY=<key> -p 9090:9090 ghcr.io/scorable/scorable-mcp:latest) or via stdio using uvx. Configure the server in your MCP client (e.g., Cursor) by pointing to the SSE endpoint or adding the stdio command with the API key.
Key features of Scorable MCP
- Exposes Scorable evaluators as MCP tools
- Supports SSE transport for network deployment
- Compatible with MCP clients like Cursor
- Tools for listing and running evaluations
- Supports evaluation by name or ID
- Includes a reference MCP client for Python
Use cases of Scorable MCP
- Evaluate and improve AI agent explanations automatically.
- Measure prompt templates for clarity and precision.
- Run coding policy adherence evaluations using policy documents.
- Use a combination of judges (collections of evaluators) for comprehensive assessment.
FAQ from Scorable MCP
How do I get an API key for Scorable MCP?
Sign up and create a key at scorable.ai/settings/api-keys, or generate a temporary key at scorable.ai/demo-user.
What transport methods does Scorable MCP support?
Scorable MCP supports SSE transport (via Docker, port 9090) and stdio transport (via uvx).
What are the main limitations of Scorable MCP?
The current implementation lacks exponential backoff, automatic retries, and request throttling. The bundled MCP client is for reference only and has no support guarantees.
Can I use Scorable MCP with Cursor?
Yes, the README provides example configurations for Cursor, using either SSE or stdio transport.
What kinds of evaluations can I run?
You can run standard evaluations by evaluator ID or name, coding policy adherence evaluations, and judges (collections of evaluators). Additional tools list all available evaluators and judges on your account.
