Danny suggested adding CI testing to catch tool changes that would break MCP server functionality:

we could have a CI job which runs Claude Code & Codex against our MCP server and do something likewhat is my coder username with a regex check. that should be pretty reliable.

This would help catch issues like the one identified in#19825 where tool schema changes could break integration with OpenAI/Anthropic.

Possible approaches:

• Run Claude/Codex against MCP server in CI
• Use test queries with regex validation
• Add e2e tests against upstream services
• Validate tools against JSON schemas

@ThomasK33