Integrate `cbioportal-mcp-qa` evals into CI and set up supporting accounts (Arize Phoenix, Claude via Bedrock)

[forus]

We want to automatically run cbioportal-mcp-qa evals as part of our CI pipeline to continuously validate application quality, correctness, and other metrics.

This requires setting up a new GitHub Actions workflow and registering external service accounts needed by the evals.

Implementation Details

1. GitHub Actions Workflow

   • Create a new CI workflow file (e.g., .github/workflows/run-evals.yml).
   • Trigger on pull requests and main branch merges.
   • Run the cbioportal-mcp-qa eval suite using existing tooling/scripts.
   • Include configurable parameters for environment (staging/production), database connection, etc.

2. Service Setup

   • Arize Phoenix
     • Register a cBioPortal account with Arize Phoenix.
     • Configure credentials/secrets in GitHub (ARIZE_API_KEY, etc.).
     • Ensure eval results are reported to Arize for tracking and monitoring.
   • Claude via Amazon Bedrock
     • Register a Claude API token (Anthropic’s Claude through Amazon Bedrock).
     • Add it to repository secrets (BEDROCK_CLAUDE_TOKEN).
     • Configure the evals to optionally leverage Claude for evaluation or summarization tasks.

3. Security

   • Store all credentials in GitHub Secrets.
   • Document how to rotate keys and tokens safely.

Acceptance Criteria

• New GitHub Actions workflow runs cbioportal-mcp-qa evals automatically on CI.
• Arize Phoenix account registered and integrated for result tracking.
• Claude (via Amazon Bedrock) token registered and configured.
• Evals produce visible pass/fail summaries in CI results (not sure how easy is that).