`--mcp` option seems not to work

[GregOnNet]

Hello everyone,

I try to use web-codegen-scorer with the Angular MCP.
After configuring it and starting the evaluation, an error is thrown.
I tried OpenAI and Anthropic models:

• gpt-5.1-no-thinking
• gpt-5.1-thinking-low
• claude-sonnet-4.5-no-thinking

That's why think the problem is not related to a specific model.

Command

export ANTHROPIC_API_KEY="..." && web-codegen-scorer eval --env=./config.mjs --model claude-sonnet-4.5-no-thinking --prompt-filter=goal-tracker --mcp

The error occurs just a few moments after running the command in the terminal

✘ [goal-tracker] Failed to generate code with AI
  AI_NoOutputGeneratedError: No output generated.
Stack: AI_NoOutputGeneratedError: No output generated.
    at get output (file:///Users/gregor/Library/pnpm/global/5/.pnpm/ai@6.0.49_zod@3.25.76/node_modules/ai/dist/index.mjs:4331:13)
    at AiSdkRunner.generateConstrained (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/codegen/ai-sdk/ai-sdk-runner.js:57:30)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
    at async AiSdkRunner.generateFiles (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/codegen/ai-sdk/ai-sdk-runner.js:68:26)
    at async generateCodeWithAI (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/orchestration/codegen.js:15:26)
    at async LocalExecutor.generateInitialFiles (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/orchestration/executors/local-executor.js:27:16)
    at async generateInitialFiles (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/orchestration/generate-initial-files.js:42:22)
    at async startEvaluationTask (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/orchestration/generate-eval-task.js:31:33)
    at async callWithTimeout (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/utils/timeout.js:17:16)
    at async evaluate (file:///Users/gregor/Library/pnpm/global/5/.pnpm/web-codegen-scorer@0.0.69_postcss@8.5.6/node_modules/web-codegen-scorer/runner/orchestration/generate.js:126:35)

Reproduction

You can find my setup here: https://github.com/GregOnNet/web-codegen-scorer-playground

I am glad to provide more information if needed.

[devversion]

Thanks for the repro! I believe this is the same as #247?

[GregOnNet]

Indeed, I hope these issues are related.

The #247 seems to focus more on the use of other thinking models.
Here it is more about adding --mcp to the command, and it is not working, - no matter which model is used.

[devversion]

@GregOnNet can you confirm it does work when you don't specify --mcp though? e.g. for GPT?

[GregOnNet]

Just checked: If I run the evaluation without --mcp and model claude-sonnet-4.5-no-thinking.

GPT models does run "forever" on my machine. 🤷