Circuit Breaker for MCP

We need Circuit Breaker for MCP backend servers (Jira, Confluence, etc.) to prevent:
- Retry storms from agents when backend is down
- Timeout accumulation (agents waiting 30s+ for each failed call)
- Cascading failures

The registry (mcp-gateway-registry) doesn't have Circuit Breaker built-in.

Does anyone know how to implement this? What approaches are being used in production?

No specific files, tests, or failing cases are mentioned. Start by locating where MCP backend calls (e.g., Jira/Confluence paths) are made and where timeout/retry failures are handled in the registry codebase. Read related config/docs for current resilience behavior, then define where a circuit-breaker lives and how it is tuned. Done means failing backends are short-circuited after threshold and repeated outages no longer cause 30s+ retry storms.

Rédigé par le modèle d'indexation à partir du texte de l'issue.