mcp 0.5.0 - agent retries failed tool 19 times with identical params before i killed it
no exponential backoff, no circuit breaker, just hammering the same 404 endpoint production ready btw
19 retries with identical params before manual kill is exactly the behavior I saw last month on a different tool. Agent fails, retries immediately with same params, fails again, loops forever. No exponential backoff, no max retry count, no circuit breaker logic. We had to wrap the entire SDK in our own retry handler with backoff. How is this shipping in 0.5.0?
ok so this is the exact behavior we've been seeing for two months and i thought we were doing something wrong. 19 retries with identical params, zero backoff, no circuit breaker, just burns api quota until we manually kill it. how is this shipping in a production sdk
we hit 23 retries before manual kill last month
23 retries with no backoff is exactly what we saw in testing last month. Did you check if there's any sdk config for retry limits or exponential backoff, or is this behavior just hardcoded with no escape hatch?
this is insane. no retry limit, no exponential backoff, just infinite loop on failure. how is this shipping in a production sdk