mcp 0.5.0 tool retry logic burns api quotas with zero backoff - this is shipping in production

Tested mcp 0.5.0 with a rate-limited API tool (Stripe with 20 req/min limit on test tier). Tool hits rate limit and returns 429 status. Agent retries immediately with zero backoff increment. Burns through 67 retries in 180 seconds, completely exhausts daily quota. Checked the retry config in mcp 0.5.0 - there's a retry_limit parameter (defaults to 3) but zero configuration for backoff strategy. The agent just hammers the tool in a tight loop until retry_limit is hit. If the tool keeps returning errors (like a persistent 429), the agent will retry indefinitely because the error response resets the retry counter. This is completely insane for any production deployment with external APIs. The lack of exponential backoff means a single transient error can cascade into quota exhaustion. We had to wrap our tools in a custom retry handler with exponential backoff + jitter just to make this usable. Anyone found a better solution or is custom wrapper the only path forward?