How to score Anthropic Claude answers with ClearRun

Steps

1. Install both

   npm install @anthropic-ai/sdk @clearrun/sdk

2. Get an answer, then score it

   import Anthropic from '@anthropic-ai/sdk';
   import { createKernelClient } from '@clearrun/sdk';
   const clearrun = createKernelClient(); // reads CLEARRUN_API_KEY
   
   const anthropic = new Anthropic();
   const prompt = 'When was the Eiffel Tower finished?';
   
   const msg = await anthropic.messages.create({
     model: process.env.ANTHROPIC_MODEL,
     max_tokens: 1024,
     messages: [{ role: 'user', content: prompt }],
   });
   const answer = msg.content
     .filter((block) => block.type === 'text')
     .map((block) => block.text)
     .join('');
   
   const { verdict } = await clearrun.evaluate({ prompt, responseText: answer });
   console.log(verdict.score, verdict.verdictId);

In Claude Desktop or Claude Code instead?

Use the MCP server. See How to add ClearRun to Claude Desktop.